Bake scale into SDPA pass1 source, remove last scalar input

This reverts commit 97d7f03032.

.mlx_typings/mlx/nn/layers/convolution.pyi

@@ -32,7 +32,6 @@ class Conv1d(Module):
     """
 
     weight: mx.array
-    bias: mx.array | None
     groups: int
     def __init__(
         self,

.mlx_typings/mlx/nn/layers/linear.pyi

@@ -40,10 +40,6 @@ class Linear(Module):
         bias (bool, optional): If set to ``False`` then the layer will
           not use a bias. Default is ``True``.
     """
-
-    weight: mx.array
-    bias: mx.array | None
-
     def __init__(self, input_dims: int, output_dims: int, bias: bool = ...) -> None: ...
     def __call__(self, x: mx.array) -> mx.array: ...
     def to_quantized(

.mlx_typings/mlx/nn/layers/normalization.pyi

@@ -88,9 +88,6 @@ class RMSNorm(Module):
         dims (int): The feature dimension of the input to normalize over
         eps (float): A small additive constant for numerical stability
     """
-
-    weight: mx.array
-
     def __init__(self, dims: int, eps: float = ...) -> None: ...
     def __call__(self, x) -> mx.array: ...
 

.mlx_typings/mlx_lm/generate.pyi

@@ -254,14 +254,7 @@ class BatchResponse:
 
     texts: List[str]
     stats: BatchStats
-    caches: Optional[List[List[Any]]]
 
-def _left_pad_prompts(prompts: Any, max_length: Optional[int] = ...) -> mx.array: ...
-def _right_pad_prompts(prompts: Any, max_length: Optional[int] = ...) -> mx.array: ...
-def _make_cache(
-    model: Any, left_padding: Any, max_kv_size: Optional[int]
-) -> List[Any]: ...
-def _merge_caches(caches: Any) -> List[Any]: ...
 @dataclass
 class Batch:
     uids: List[int]
@@ -270,71 +263,39 @@ class Batch:
     max_tokens: List[int]
     num_tokens: List[int]
     cache: List[Any]
-    samplers: List[Any]
-    logits_processors: List[Any]
-    tokens: List[mx.array]
-    def __len__(self) -> int: ...
-    def filter(self, keep_idx: List[int]) -> None: ...
-    def extend(self, other: "Batch") -> None: ...
-    def extract_cache(self, idx: int) -> List[Any]: ...
+    def __len__(self):  # -> int:
+        ...
+    def filter(self, keep_idx: List[int]):  # -> None:
+        ...
+    def extend(self, other):  # -> None:
+        ...
 
 class BatchGenerator:
-    model: Any
-    max_kv_size: Optional[int]
-    prefill_step_size: int
-    unprocessed_prompts: List[Any]
-    active_batch: Optional[Batch]
-    prompt_progress_callback: Callable[[List[Tuple[int, int, int]]], None]
-    _stats: BatchStats
-
     @dataclass
     class Response:
         uid: int
         token: int
         logprobs: mx.array
         finish_reason: Optional[str]
-        prompt_cache: Any
 
     def __init__(
         self,
-        model: Any,
+        model,
         max_tokens: int = ...,
-        stop_tokens: Optional[set[int]] = ...,
+        stop_tokens: Optional[set] = ...,
         sampler: Optional[Callable[[mx.array], mx.array]] = ...,
-        logits_processors: Optional[
-            List[Callable[[mx.array, mx.array], mx.array]]
-        ] = ...,
         completion_batch_size: int = ...,
         prefill_batch_size: int = ...,
         prefill_step_size: int = ...,
-        prompt_progress_callback: Optional[
-            Callable[[List[Tuple[int, int, int]]], None]
-        ] = ...,
-        max_kv_size: Optional[int] = ...,
     ) -> None: ...
-    def close(self) -> None: ...
     def insert(
-        self,
-        prompts: Any,
-        max_tokens: Union[List[int], int, None] = ...,
-        caches: Any = ...,
-        samplers: Optional[List[Any]] = ...,
-        logits_processors: Optional[List[Any]] = ...,
-    ) -> List[int]: ...
-    def remove(
-        self, uids: List[int], return_prompt_caches: bool = ...
-    ) -> Optional[dict[int, List[Any]]]: ...
-    def stats(self) -> BatchStats: ...
-    def next(self) -> List[Response]: ...
-    def _process_prompts(self, prompts: List[Any]) -> Batch: ...
-    def _step(
-        self,
-        input_tokens: mx.array,
-        prompt_cache: List[Any],
-        samplers: Optional[List[Any]],
-        logits_processors: Optional[List[Any]],
-        tokens: List[mx.array],
-    ) -> Tuple[mx.array, List[mx.array]]: ...
+        self, prompts, max_tokens: Union[List[int], int, None] = ...
+    ):  # -> list[Any]:
+        ...
+    def stats(self):  # -> BatchStats:
+        ...
+    def next(self):  # -> list[Any]:
+        ...
 
 def batch_generate(
     model,

.mlx_typings/mlx_lm/models/cache.pyi

@@ -170,15 +170,7 @@ class KVCache(_BaseCache):
 
 class RotatingKVCache(_BaseCache):
     step = ...
-    keys: mx.array | None
-    values: mx.array | None
-    keep: int
-    max_size: int
-    _idx: int
     def __init__(self, max_size, keep=...) -> None: ...
-    def _trim(
-        self, trim_size: int, v: mx.array, append: mx.array | None = ...
-    ) -> mx.array: ...
     def update_and_fetch(
         self, keys, values
     ):  # -> tuple[array | Any, array | Any] | tuple[array | Any, array | Any | None]:

.mlx_typings/mlx_lm/models/nemotron_h.pyi

@@ -1,154 +0,0 @@
-from dataclasses import dataclass
-from typing import Any, List, Optional, Tuple
-
-import mlx.core as mx
-import mlx.nn as nn
-
-from .cache import ArraysCache, KVCache
-from .switch_layers import SwitchMLP
-
-@dataclass
-class ModelArgs:
-    model_type: str
-    vocab_size: int
-    hidden_size: int
-    intermediate_size: int
-    num_hidden_layers: int
-    max_position_embeddings: int
-    num_attention_heads: int
-    num_key_value_heads: int
-    attention_bias: bool
-    mamba_num_heads: int
-    mamba_head_dim: int
-    mamba_proj_bias: bool
-    ssm_state_size: int
-    conv_kernel: int
-    n_groups: int
-    mlp_bias: bool
-    layer_norm_epsilon: float
-    use_bias: bool
-    use_conv_bias: bool
-    hybrid_override_pattern: List[str]
-    head_dim: Optional[int]
-    moe_intermediate_size: Optional[int]
-    moe_shared_expert_intermediate_size: Optional[int]
-    n_group: Optional[int]
-    n_routed_experts: Optional[int]
-    n_shared_experts: Optional[int]
-    topk_group: Optional[int]
-    num_experts_per_tok: Optional[int]
-    norm_topk_prob: Optional[bool]
-    routed_scaling_factor: Optional[float]
-    time_step_limit: Optional[Tuple[float, float]]
-    time_step_min: Optional[float]
-    time_step_max: Optional[float]
-
-    @classmethod
-    def from_dict(cls, params: dict[str, Any]) -> ModelArgs: ...
-    def __post_init__(self) -> None: ...
-
-class NemotronHMamba2Mixer(nn.Module):
-    num_heads: int
-    hidden_size: int
-    ssm_state_size: int
-    conv_kernel_size: int
-    intermediate_size: int
-    n_groups: int
-    head_dim: int
-    conv_dim: int
-    conv1d: nn.Conv1d
-    in_proj: nn.Linear
-    dt_bias: mx.array
-    A_log: mx.array
-    D: mx.array
-    norm: nn.RMSNorm
-    heads_per_group: int
-    out_proj: nn.Linear
-
-    def __init__(self, args: ModelArgs) -> None: ...
-    def __call__(
-        self,
-        hidden_states: mx.array,
-        mask: Optional[mx.array],
-        cache: Optional[ArraysCache] = None,
-    ) -> mx.array: ...
-
-class NemotronHAttention(nn.Module):
-    hidden_size: int
-    num_heads: int
-    head_dim: int
-    num_key_value_heads: int
-    scale: float
-    q_proj: nn.Linear
-    k_proj: nn.Linear
-    v_proj: nn.Linear
-    o_proj: nn.Linear
-
-    def __init__(self, args: ModelArgs) -> None: ...
-    def __call__(
-        self,
-        x: mx.array,
-        mask: Optional[mx.array] = None,
-        cache: Optional[KVCache] = None,
-    ) -> mx.array: ...
-
-class NemotronHMLP(nn.Module):
-    up_proj: nn.Linear
-    down_proj: nn.Linear
-
-    def __init__(
-        self, args: ModelArgs, intermediate_size: Optional[int] = None
-    ) -> None: ...
-    def __call__(self, x: mx.array) -> mx.array: ...
-
-class NemotronHMoE(nn.Module):
-    num_experts_per_tok: int
-    switch_mlp: SwitchMLP
-    shared_experts: NemotronHMLP
-
-    def __init__(self, config: ModelArgs) -> None: ...
-    def __call__(self, x: mx.array) -> mx.array: ...
-
-class NemotronHBlock(nn.Module):
-    block_type: str
-    norm: nn.RMSNorm
-    mixer: NemotronHMamba2Mixer | NemotronHAttention | NemotronHMLP | NemotronHMoE
-
-    def __init__(self, args: ModelArgs, block_type: str) -> None: ...
-    def __call__(
-        self,
-        x: mx.array,
-        mask: Optional[mx.array] = None,
-        cache: Optional[Any] = None,
-    ) -> mx.array: ...
-
-class NemotronHModel(nn.Module):
-    embeddings: nn.Embedding
-    layers: list[NemotronHBlock]
-    norm_f: nn.RMSNorm
-    fa_idx: int
-    ssm_idx: int
-
-    def __init__(self, args: ModelArgs) -> None: ...
-    def __call__(
-        self,
-        inputs: mx.array,
-        cache: Optional[Any] = None,
-    ) -> mx.array: ...
-
-class Model(nn.Module):
-    args: ModelArgs
-    backbone: NemotronHModel
-    lm_head: nn.Linear
-    model_type: str
-
-    def __init__(self, args: ModelArgs) -> None: ...
-    def __call__(
-        self,
-        inputs: mx.array,
-        cache: Optional[Any] = None,
-    ) -> mx.array: ...
-    @property
-    def layers(self) -> list[NemotronHBlock]: ...
-    def make_cache(self) -> list[ArraysCache | KVCache]: ...
-    def sanitize(self, weights: dict[str, Any]) -> dict[str, Any]: ...

.mlx_typings/mlx_lm/models/qwen3_next.pyi

@@ -5,7 +5,6 @@ from typing import Any, Optional
 import mlx.core as mx
 import mlx.nn as nn
 
-from .cache import ArraysCache, KVCache
 from .switch_layers import SwitchGLU
 
 class Qwen3NextRMSNormGated(nn.Module):
@@ -100,8 +99,6 @@ class Qwen3NextModel(nn.Module):
     embed_tokens: nn.Embedding
     layers: list[Qwen3NextDecoderLayer]
     norm: nn.RMSNorm
-    ssm_idx: int
-    fa_idx: int
 
     def __init__(self, args: Any) -> None: ...
     def __call__(
@@ -124,4 +121,3 @@ class Model(nn.Module):
     def sanitize(self, weights: dict[str, Any]) -> dict[str, Any]: ...
     @property
     def layers(self) -> list[Qwen3NextDecoderLayer]: ...
-    def make_cache(self) -> list[ArraysCache | KVCache]: ...

.mlx_typings/mlx_lm/models/switch_layers.pyi

@@ -73,9 +73,6 @@ class SwitchGLU(nn.Module):
     def __call__(self, x, indices) -> mx.array: ...
 
 class SwitchMLP(nn.Module):
-    fc1: SwitchLinear
-    fc2: SwitchLinear
-
     def __init__(
         self,
         input_dims: int,

.mlx_typings/mlx_lm/tokenizer_utils.pyi

@@ -39,11 +39,11 @@ class StreamingDetokenizer:
     """
 
     __slots__ = ...
-    def reset(self) -> None: ...
-    def add_token(self, token: int) -> None: ...
-    def finalize(self) -> None: ...
+    def reset(self): ...
+    def add_token(self, token): ...
+    def finalize(self): ...
     @property
-    def last_segment(self) -> str:
+    def last_segment(self):
         """Return the last segment of readable text since last time this property was accessed."""
 
 class NaiveStreamingDetokenizer(StreamingDetokenizer):

Cargo.toml

@@ -1,11 +1,6 @@
 [workspace]
 resolver = "3"
-members = [
-  "rust/networking",
-  "rust/exo_pyo3_bindings",
-  "rust/util",
-  "rust/babblerd",
-]
+members = ["rust/networking", "rust/exo_pyo3_bindings", "rust/util"]
 
 [workspace.package]
 version = "0.0.1"
@@ -48,6 +43,7 @@ log = "0.4"
 
 # networking
 libp2p = "0.56"
+libp2p-tcp = "0.44"
 
 [workspace.lints.rust]
 static_mut_refs = "warn"      # Or use "warn" instead of deny

bench/eval_configs/models.toml

@@ -1,292 +0,0 @@
-# Model evaluation configurations for exo_eval.
-#
-# Each [[model]] entry uses `patterns` — a list of substrings matched
-# against the model_id. First matching entry wins.
-#
-# Required fields:
-#   name, patterns, reasoning
-#
-# Optional per-model overrides (CLI flags take priority over these):
-#   temperature, top_p, max_tokens, reasoning_effort
-#
-# Fallback defaults (when no per-model config):
-#   reasoning:     temperature=1.0, max_tokens=131072, reasoning_effort="high"
-#   non-reasoning: temperature=0.0, max_tokens=16384
-#
-# All per-model values below are sourced from official model cards,
-# generation_config.json files, and vendor documentation.
-
-# ─── Qwen3.5 (Feb 2026) ─────────────────────────────────────────────
-# Source: HuggingFace model cards (Qwen/Qwen3.5-*)
-# 35B-A3B thinking general: temp=1.0, top_p=0.95, top_k=20
-# 397B thinking: temp=0.6, top_p=0.95, top_k=20
-# Non-thinking: temp=0.7, top_p=0.8, top_k=20
-# max_tokens: 32768 general, 81920 for complex math/code
-
-[[model]]
-name = "Qwen3.5 2B"
-patterns = ["Qwen3.5-2B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 81920
-
-[[model]]
-name = "Qwen3.5 9B"
-patterns = ["Qwen3.5-9B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 81920
-
-[[model]]
-name = "Qwen3.5 27B"
-patterns = ["Qwen3.5-27B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 81920
-
-[[model]]
-name = "Qwen3.5 35B A3B"
-patterns = ["Qwen3.5-35B-A3B"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-max_tokens = 81920
-
-[[model]]
-name = "Qwen3.5 122B A10B"
-patterns = ["Qwen3.5-122B-A10B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 81920
-
-[[model]]
-name = "Qwen3.5 397B A17B"
-patterns = ["Qwen3.5-397B-A17B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 81920
-
-# ─── Qwen3 (Apr 2025) ───────────────────────────────────────────────
-# Source: HuggingFace model cards (Qwen/Qwen3-*)
-# Thinking: temp=0.6, top_p=0.95, top_k=20
-# Non-thinking: temp=0.7, top_p=0.8, top_k=20
-# max_tokens: 32768 general, 38912 for complex math/code
-
-[[model]]
-name = "Qwen3 0.6B"
-patterns = ["Qwen3-0.6B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 38912
-
-[[model]]
-name = "Qwen3 30B A3B"
-patterns = ["Qwen3-30B-A3B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 38912
-
-[[model]]
-name = "Qwen3 235B A22B"
-patterns = ["Qwen3-235B-A22B"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 38912
-
-[[model]]
-name = "Qwen3 Next 80B Thinking"
-patterns = ["Qwen3-Next-80B-A3B-Thinking"]
-reasoning = true
-temperature = 0.6
-top_p = 0.95
-max_tokens = 38912
-
-[[model]]
-name = "Qwen3 Next 80B Instruct"
-patterns = ["Qwen3-Next-80B-A3B-Instruct"]
-reasoning = false
-temperature = 0.7
-top_p = 0.8
-max_tokens = 16384
-
-[[model]]
-name = "Qwen3 Coder 480B"
-patterns = ["Qwen3-Coder-480B"]
-reasoning = false
-temperature = 0.7
-top_p = 0.8
-max_tokens = 16384
-
-[[model]]
-name = "Qwen3 Coder Next"
-patterns = ["Qwen3-Coder-Next"]
-reasoning = false
-temperature = 0.7
-top_p = 0.8
-max_tokens = 16384
-
-# ─── GPT-OSS (OpenAI) ───────────────────────────────────────────────
-# Source: OpenAI GitHub README + HuggingFace discussion #21
-# temp=1.0, top_p=1.0, NO top_k, NO repetition_penalty
-# reasoning_effort supported: low/medium/high
-
-[[model]]
-name = "GPT-OSS 20B"
-patterns = ["gpt-oss-20b"]
-reasoning = true
-temperature = 1.0
-top_p = 1.0
-
-[[model]]
-name = "GPT-OSS 120B"
-patterns = ["gpt-oss-120b"]
-reasoning = true
-temperature = 1.0
-top_p = 1.0
-
-# ─── DeepSeek ────────────────────────────────────────────────────────
-# Source: https://api-docs.deepseek.com/quick_start/parameter_settings
-# Coding/Math: temp=0.0, General: temp=1.3, Creative: temp=1.5
-# NOTE: DeepSeek API applies nonlinear temp mapping. These are API values.
-# When running model directly: API temp 1.0 = model temp 0.3
-# We use temp=0.0 for eval (coding/math focus).
-
-[[model]]
-name = "DeepSeek V3.1"
-patterns = ["DeepSeek-V3.1"]
-reasoning = true
-temperature = 0.0
-
-# ─── GLM (ZhipuAI / THUDM) ──────────────────────────────────────────
-# Source: HuggingFace model cards + generation_config.json + docs.z.ai
-# GLM 4.5+: temp=1.0, top_p=0.95
-# Reasoning tasks: 131072 max_tokens; coding/SWE tasks: temp=0.7
-
-[[model]]
-name = "GLM-5"
-patterns = ["GLM-5"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-max_tokens = 131072
-
-[[model]]
-name = "GLM 4.5 Air"
-patterns = ["GLM-4.5-Air"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-
-[[model]]
-name = "GLM 4.7"
-patterns = ["GLM-4.7-"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-max_tokens = 131072
-# Note: matches both GLM-4.7 and GLM-4.7-Flash
-
-# ─── Kimi (Moonshot AI) ─────────────────────────────────────────────
-# Source: HuggingFace model cards (moonshotai/Kimi-K2-*)
-# K2-Instruct: temp=0.6
-# K2-Thinking: temp=1.0, max_length=262144
-# K2.5: thinking temp=1.0, top_p=0.95; instant temp=0.6, top_p=0.95
-
-[[model]]
-name = "Kimi K2 Thinking"
-patterns = ["Kimi-K2-Thinking"]
-reasoning = true
-temperature = 1.0
-max_tokens = 131072
-
-[[model]]
-name = "Kimi K2.5"
-patterns = ["Kimi-K2.5"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-max_tokens = 131072
-
-[[model]]
-name = "Kimi K2 Instruct"
-patterns = ["Kimi-K2-Instruct"]
-reasoning = false
-temperature = 0.6
-
-# ─── MiniMax ─────────────────────────────────────────────────────────
-# Source: HuggingFace model cards + generation_config.json
-# All models: temp=1.0, top_p=0.95, top_k=40
-
-[[model]]
-name = "MiniMax M2.5"
-patterns = ["MiniMax-M2.5"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-
-[[model]]
-name = "MiniMax M2.1"
-patterns = ["MiniMax-M2.1"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-
-# ─── Step (StepFun) ─────────────────────────────────────────────────
-# Source: HuggingFace model card (stepfun-ai/Step-3.5-Flash)
-# Reasoning: temp=1.0, top_p=0.95
-# General chat: temp=0.6, top_p=0.95
-# We use reasoning settings for eval.
-
-[[model]]
-name = "Step 3.5 Flash"
-patterns = ["Step-3.5-Flash"]
-reasoning = true
-temperature = 1.0
-top_p = 0.95
-
-# ─── Llama (Meta) ───────────────────────────────────────────────────
-# Source: generation_config.json + meta-llama/llama-models generation.py
-# All variants: temp=0.6, top_p=0.9
-
-[[model]]
-name = "Llama 3.2 1B"
-patterns = ["Llama-3.2-1B"]
-reasoning = false
-temperature = 0.6
-top_p = 0.9
-
-[[model]]
-name = "Llama 3.2 3B"
-patterns = ["Llama-3.2-3B"]
-reasoning = false
-temperature = 0.6
-top_p = 0.9
-
-[[model]]
-name = "Llama 3.1 8B"
-patterns = ["Llama-3.1-8B", "Meta-Llama-3.1-8B"]
-reasoning = false
-temperature = 0.6
-top_p = 0.9
-
-[[model]]
-name = "Llama 3.1 70B"
-patterns = ["Llama-3.1-70B", "Meta-Llama-3.1-70B"]
-reasoning = false
-temperature = 0.6
-top_p = 0.9
-
-[[model]]
-name = "Llama 3.3 70B"
-patterns = ["Llama-3.3-70B", "llama-3.3-70b"]
-reasoning = false
-temperature = 0.6
-top_p = 0.9

bench/exo_bench.py

@@ -24,7 +24,6 @@ import json
 import sys
 import time
 from collections.abc import Callable
-from concurrent.futures import ThreadPoolExecutor, as_completed
 from pathlib import Path
 from statistics import mean
 from typing import Any
@@ -253,12 +252,6 @@ def main() -> int:
     ap.add_argument(
         "--repeat", type=int, default=1, help="Repetitions per (pp,tg) pair."
     )
-    ap.add_argument(
-        "--concurrency",
-        nargs="+",
-        default=["1"],
-        help="Concurrency levels (ints). Accepts commas. E.g. --concurrency 1,2,4,8. Default 1.",
-    )
     ap.add_argument(
         "--warmup",
         type=int,
@@ -289,10 +282,6 @@ def main() -> int:
     if args.repeat <= 0:
         logger.error("--repeat must be >= 1")
         return 2
-    concurrency_list = parse_int_list(args.concurrency)
-    if not concurrency_list or any(c <= 0 for c in concurrency_list):
-        logger.error("--concurrency values must be >= 1")
-        return 2
 
     # Log pairing mode
     use_combinations = args.all_combinations or len(pp_list) != len(tg_list)
@@ -304,9 +293,7 @@ def main() -> int:
         logger.info(f"pp/tg mode: tandem (zip) - {len(pp_list)} pairs")
 
     client = ExoClient(args.host, args.port, timeout_s=args.timeout)
-    short_id, full_model_id = resolve_model_short_id(
-        client, args.model, force_download=args.force_download
-    )
+    short_id, full_model_id = resolve_model_short_id(client, args.model)
 
     tokenizer = load_tokenizer_for_bench(full_model_id)
     if tokenizer is None:
@@ -405,131 +392,52 @@ def main() -> int:
                 pp_tg_pairs = list(zip(pp_list, tg_list, strict=True))
 
             for pp, tg in pp_tg_pairs:
-                for concurrency in concurrency_list:
-                    logger.info(f"--- pp={pp} tg={tg} concurrency={concurrency} ---")
-                    runs: list[dict[str, Any]] = []
-                    for r in range(args.repeat):
-                        time.sleep(3)
-
-                        if concurrency <= 1:
-                            # Sequential: single request
-                            try:
-                                row, actual_pp_tokens = run_one_completion(
-                                    client, full_model_id, pp, tg, prompt_sizer
-                                )
-                            except Exception as e:
-                                logger.error(e)
-                                continue
-                            row.update(
-                                {
-                                    "model_short_id": short_id,
-                                    "model_id": full_model_id,
-                                    "placement_sharding": sharding,
-                                    "placement_instance_meta": instance_meta,
-                                    "placement_nodes": n_nodes,
-                                    "instance_id": instance_id,
-                                    "pp_tokens": actual_pp_tokens,
-                                    "tg": tg,
-                                    "repeat_index": r,
-                                    "concurrency": 1,
-                                    **(
-                                        {"download_duration_s": download_duration_s}
-                                        if download_duration_s is not None
-                                        else {}
-                                    ),
-                                }
-                            )
-                            runs.append(row)
-                            all_rows.append(row)
-                        else:
-                            # Concurrent: fire N requests in parallel
-                            # Each thread gets its own ExoClient (separate HTTP connection)
-                            batch_results: list[tuple[dict[str, Any], int]] = []
-                            batch_errors = 0
-
-                            def _run_concurrent(
-                                idx: int, *, _pp: int = pp, _tg: int = tg
-                            ) -> tuple[dict[str, Any], int]:
-                                c = ExoClient(
-                                    args.host, args.port, timeout_s=args.timeout
-                                )
-                                return run_one_completion(
-                                    c, full_model_id, _pp, _tg, prompt_sizer
-                                )
-
-                            with ThreadPoolExecutor(max_workers=concurrency) as pool:
-                                futures = {
-                                    pool.submit(_run_concurrent, i): i
-                                    for i in range(concurrency)
-                                }
-                                for fut in as_completed(futures):
-                                    try:
-                                        batch_results.append(fut.result())
-                                    except Exception as e:
-                                        logger.error(f"Concurrent request failed: {e}")
-                                        batch_errors += 1
-
-                            for idx, (row, actual_pp_tokens) in enumerate(
-                                batch_results
-                            ):
-                                row.update(
-                                    {
-                                        "model_short_id": short_id,
-                                        "model_id": full_model_id,
-                                        "placement_sharding": sharding,
-                                        "placement_instance_meta": instance_meta,
-                                        "placement_nodes": n_nodes,
-                                        "instance_id": instance_id,
-                                        "pp_tokens": actual_pp_tokens,
-                                        "tg": tg,
-                                        "repeat_index": r,
-                                        "concurrency": concurrency,
-                                        "concurrent_index": idx,
-                                        **(
-                                            {"download_duration_s": download_duration_s}
-                                            if download_duration_s is not None
-                                            else {}
-                                        ),
-                                    }
-                                )
-                                runs.append(row)
-                                all_rows.append(row)
-
-                            if batch_results:
-                                valid_gen_tps = [
-                                    x["stats"]["generation_tps"]
-                                    for x, _ in batch_results
-                                    if x["stats"]["generation_tps"] > 0
-                                ]
-                                agg_gen_tps = (
-                                    mean(valid_gen_tps) if valid_gen_tps else 0.0
-                                )
-                                gen_tps = agg_gen_tps / concurrency
-                                logger.info(
-                                    f"[concurrent {concurrency}x]  "
-                                    f"agg_gen_tps={agg_gen_tps:.2f}  "
-                                    f"gen_tps={gen_tps:.2f}  "
-                                    f"errors={batch_errors}"
-                                )
-
-                    if runs:
-                        prompt_tps = mean(x["stats"]["prompt_tps"] for x in runs)
-                        gen_tps = mean(
-                            x["stats"]["generation_tps"] / x["concurrency"]
-                            for x in runs
-                        )
-                        ptok = mean(x["stats"]["prompt_tokens"] for x in runs)
-                        gtok = mean(x["stats"]["generation_tokens"] for x in runs)
-                        peak = mean(
-                            x["stats"]["peak_memory_usage"]["inBytes"] for x in runs
+                runs: list[dict[str, Any]] = []
+                for r in range(args.repeat):
+                    time.sleep(3)
+                    try:
+                        row, actual_pp_tokens = run_one_completion(
+                            client, full_model_id, pp, tg, prompt_sizer
                         )
+                    except Exception as e:
+                        logger.error(e)
+                        continue
+                    row.update(
+                        {
+                            "model_short_id": short_id,
+                            "model_id": full_model_id,
+                            "placement_sharding": sharding,
+                            "placement_instance_meta": instance_meta,
+                            "placement_nodes": n_nodes,
+                            "instance_id": instance_id,
+                            "pp_tokens": actual_pp_tokens,
+                            "tg": tg,
+                            "repeat_index": r,
+                            **(
+                                {"download_duration_s": download_duration_s}
+                                if download_duration_s is not None
+                                else {}
+                            ),
+                        }
+                    )
+                    runs.append(row)
+                    all_rows.append(row)
 
-                        logger.info(
-                            f"prompt_tps={prompt_tps:.2f} gen_tps={gen_tps:.2f}    "
-                            f"prompt_tokens={ptok} gen_tokens={gtok}    "
-                            f"peak_memory={format_peak_memory(peak)}\n"
-                        )
-                    time.sleep(2)
+                if runs:
+                    prompt_tps = mean(x["stats"]["prompt_tps"] for x in runs)
+                    gen_tps = mean(x["stats"]["generation_tps"] for x in runs)
+                    ptok = mean(x["stats"]["prompt_tokens"] for x in runs)
+                    gtok = mean(x["stats"]["generation_tokens"] for x in runs)
+                    peak = mean(
+                        x["stats"]["peak_memory_usage"]["inBytes"] for x in runs
+                    )
+
+                    logger.info(
+                        f"prompt_tps={prompt_tps:.2f} gen_tps={gen_tps:.2f}    "
+                        f"prompt_tokens={ptok} gen_tokens={gtok}    "
+                        f"peak_memory={format_peak_memory(peak)}\n"
+                    )
+                time.sleep(2)
         finally:
             try:
                 client.request_json("DELETE", f"/instance/{instance_id}")

bench/harness.py

@@ -165,9 +165,7 @@ def wait_for_instance_gone(
     raise TimeoutError(f"Instance {instance_id} did not get deleted within {timeout=}")
 
 
-def resolve_model_short_id(
-    client: ExoClient, model_arg: str, *, force_download: bool = False
-) -> tuple[str, str]:
+def resolve_model_short_id(client: ExoClient, model_arg: str) -> tuple[str, str]:
     models = client.request_json("GET", "/models") or {}
     data = models.get("data") or []
 
@@ -183,16 +181,6 @@ def resolve_model_short_id(
             full_id = str(m["hugging_face_id"])
             return short_id, full_id
 
-    if force_download and "/" in model_arg:
-        logger.info(f"Model not in /models, adding from HuggingFace: {model_arg}")
-        result = client.request_json(
-            "POST", "/models/add", body={"model_id": model_arg}
-        )
-        if result:
-            short_id = str(result.get("name") or model_arg.rsplit("/", 1)[-1])
-            full_id = str(result.get("hugging_face_id") or model_arg)
-            return short_id, full_id
-
     raise ValueError(f"Model not found in /models: {model_arg}")
 
 
@@ -457,11 +445,6 @@ def add_common_instance_args(ap: argparse.ArgumentParser) -> None:
         "--port", type=int, default=int(os.environ.get("EXO_PORT", "52415"))
     )
     ap.add_argument("--model", required=True, help="Model short id or huggingface id")
-    ap.add_argument(
-        "--force-download",
-        action="store_true",
-        help="If model not in /models, add it from HuggingFace via exo and download.",
-    )
     ap.add_argument(
         "--max-nodes",
         type=int,

bench/parallel_requests.py

@@ -1,255 +0,0 @@
-# type: ignore
-import argparse
-import asyncio
-import sys
-import termios
-import time
-import tty
-
-import aiohttp
-
-NUM_REQUESTS = 10
-BASE_URL = ""
-
-QUESTIONS = [
-    "What is the capital of Australia?",
-    "How many bones are in the human body?",
-    "What year did World War II end?",
-    "What is the speed of light in meters per second?",
-    "Who wrote Romeo and Juliet?",
-    "What is the chemical formula for water?",
-    "How many planets are in our solar system?",
-    "What is the largest ocean on Earth?",
-    "Who painted the Mona Lisa?",
-    "What is the boiling point of water in Celsius?",
-]
-
-
-def write(s: str) -> None:
-    sys.stdout.write(s)
-
-
-# ---------------------------------------------------------------------------
-# Model picker (same style as exo_eval)
-# ---------------------------------------------------------------------------
-
-
-def fetch_models() -> list[str]:
-    import json
-    import urllib.request
-
-    with urllib.request.urlopen(f"{BASE_URL}/state") as resp:
-        data = json.loads(resp.read())
-    model_ids: set[str] = set()
-    for instance in data.get("instances", {}).values():
-        for variant in instance.values():
-            sa = variant.get("shardAssignments", {})
-            model_id = sa.get("modelId")
-            if model_id:
-                model_ids.add(model_id)
-    return sorted(model_ids)
-
-
-def pick_model() -> str | None:
-    models = fetch_models()
-    if not models:
-        print("No models found.")
-        return None
-
-    cursor = 0
-    total_lines = len(models) + 4
-
-    def render(first: bool = False) -> None:
-        if not first:
-            write(f"\033[{total_lines}A")
-        write("\033[J")
-        write("\033[1mSelect model\033[0m (up/down, enter confirm, q quit)\r\n\r\n")
-        for i, model in enumerate(models):
-            line = f"  {'>' if i == cursor else ' '} {model}"
-            write(f"\033[7m{line}\033[0m\r\n" if i == cursor else f"{line}\r\n")
-        write("\r\n")
-        sys.stdout.flush()
-
-    fd = sys.stdin.fileno()
-    old = termios.tcgetattr(fd)
-    try:
-        tty.setraw(fd)
-        write("\033[?25l")
-        render(first=True)
-        while True:
-            ch = sys.stdin.read(1)
-            if ch in ("q", "\x03"):
-                write("\033[?25h\033[0m\r\n")
-                return None
-            elif ch in ("\r", "\n"):
-                break
-            elif ch == "\x1b":
-                seq = sys.stdin.read(2)
-                if seq == "[A":
-                    cursor = (cursor - 1) % len(models)
-                elif seq == "[B":
-                    cursor = (cursor + 1) % len(models)
-            render()
-    finally:
-        termios.tcsetattr(fd, termios.TCSADRAIN, old)
-        write(f"\033[{total_lines}A\033[J")  # clear picker UI
-        write("\033[?25h\033[0m")
-        sys.stdout.flush()
-
-    return models[cursor]
-
-
-# ---------------------------------------------------------------------------
-# Parallel requests
-# ---------------------------------------------------------------------------
-
-statuses: list[str] = []
-times: list[str] = []
-previews: list[str] = []
-tokens: list[str] = []
-full_responses: list[dict | None] = []
-total_lines = 0
-start_time: float = 0
-selected_model: str = ""
-
-
-def render_progress(first: bool = False) -> None:
-    if not first:
-        write(f"\033[{total_lines}A")
-    write("\033[J")
-    elapsed = time.monotonic() - start_time if start_time else 0
-    done = sum(1 for s in statuses if s == "done")
-    write(
-        f"\033[1m{selected_model}\033[0m  [{done}/{NUM_REQUESTS}]  {elapsed:.1f}s\r\n\r\n"
-    )
-
-    for i in range(NUM_REQUESTS):
-        q = QUESTIONS[i % len(QUESTIONS)]
-        status = statuses[i]
-        if status == "pending":
-            color = "\033[33m"  # yellow
-        elif status == "running":
-            color = "\033[36m"  # cyan
-        elif status == "done":
-            color = "\033[32m"  # green
-        else:
-            color = "\033[31m"  # red
-        write(
-            f"  {i:>2}  {color}{status:<8}\033[0m  {times[i]:>6}  {tokens[i]:>5}tok  {q[:40]:<40}  {previews[i][:50]}\r\n"
-        )
-
-    write("\r\n")
-    sys.stdout.flush()
-
-
-async def send_request(
-    session: aiohttp.ClientSession, i: int, lock: asyncio.Lock
-) -> None:
-    payload = {
-        "model": selected_model,
-        "messages": [{"role": "user", "content": QUESTIONS[i % len(QUESTIONS)]}],
-        "max_tokens": 1024,
-    }
-    statuses[i] = "running"
-    async with lock:
-        render_progress()
-    t0 = time.monotonic()
-    try:
-        async with session.post(
-            f"{BASE_URL}/v1/chat/completions", json=payload
-        ) as resp:
-            data = await resp.json()
-            elapsed = time.monotonic() - t0
-            full_responses[i] = data
-            times[i] = f"{elapsed:.1f}s"
-            if resp.status == 200:
-                choice = data["choices"][0]
-                msg = choice["message"]
-                content = msg.get("content", "")
-                previews[i] = content[:50].replace("\n", " ") or "(empty)"
-                if "usage" in data:
-                    tokens[i] = str(data["usage"].get("total_tokens", ""))
-                statuses[i] = "done"
-            else:
-                statuses[i] = f"err:{resp.status}"
-                previews[i] = str(data.get("error", {}).get("message", ""))[:50]
-    except Exception as e:
-        elapsed = time.monotonic() - t0
-        times[i] = f"{elapsed:.1f}s"
-        statuses[i] = "error"
-        previews[i] = str(e)[:50]
-    async with lock:
-        render_progress()
-
-
-async def run_requests(print_stdout: bool = False) -> None:
-    global start_time, total_lines, statuses, times, previews, tokens, full_responses
-
-    statuses = ["pending"] * NUM_REQUESTS
-    times = ["-"] * NUM_REQUESTS
-    previews = ["-"] * NUM_REQUESTS
-    tokens = ["-"] * NUM_REQUESTS
-    full_responses = [None] * NUM_REQUESTS
-    total_lines = NUM_REQUESTS + 4
-
-    write("\033[?25l")  # hide cursor
-    start_time = time.monotonic()
-    render_progress(first=True)
-    lock = asyncio.Lock()
-    try:
-        async with aiohttp.ClientSession() as session:
-            tasks = [send_request(session, i, lock) for i in range(NUM_REQUESTS)]
-            await asyncio.gather(*tasks)
-        total = time.monotonic() - start_time
-        write(
-            f"\033[1m=== All {NUM_REQUESTS} requests done in {total:.1f}s ===\033[0m\r\n\r\n"
-        )
-
-        if print_stdout:
-            for i in range(NUM_REQUESTS):
-                data = full_responses[i]
-                if not data or "choices" not in data:
-                    continue
-                choice = data["choices"][0]
-                msg = choice["message"]
-                q = QUESTIONS[i % len(QUESTIONS)]
-                write(f"\033[1m--- #{i}: {q} ---\033[0m\r\n")
-                if msg.get("reasoning_content"):
-                    write(f"\033[2m[Thinking]: {msg['reasoning_content']}\033[0m\r\n")
-                write(f"{msg.get('content', '')}\r\n")
-                if "usage" in data:
-                    u = data["usage"]
-                    write(
-                        f"\033[2m[Usage: prompt={u.get('prompt_tokens')}, "
-                        f"completion={u.get('completion_tokens')}, "
-                        f"total={u.get('total_tokens')}]\033[0m\r\n"
-                    )
-                write("\r\n")
-    finally:
-        write("\033[?25h")  # show cursor
-        sys.stdout.flush()
-
-
-def main() -> None:
-    global selected_model, BASE_URL
-    parser = argparse.ArgumentParser(
-        description="Send parallel requests to an exo cluster"
-    )
-    parser.add_argument(
-        "--host", required=True, help="Hostname of the exo node (e.g. s1)"
-    )
-    parser.add_argument("--port", type=int, default=52415, help="Port (default: 52415)")
-    parser.add_argument(
-        "--stdout", action="store_true", help="Print full responses after completion"
-    )
-    args = parser.parse_args()
-    BASE_URL = f"http://{args.host}:{args.port}"
-    model = pick_model()
-    if not model:
-        return
-    selected_model = model
-    asyncio.run(run_requests(print_stdout=args.stdout))
-
-
-if __name__ == "__main__":
-    main()

bench/pyproject.toml

@@ -11,11 +11,6 @@ dependencies = [
   "tiktoken>=0.12.0",
   "jinja2>=3.1.0",
   "protobuf>=5.29.0",
-  "datasets>=2.0.0",
-  "math-verify>=0.7.0",
-  "lm-eval[api,math]>=0.4.0",
-  "human-eval>=1.0.3",
-  "numpy>=1.24.0",
 ]
 
 [build-system]

bench/test_mlx_bandwidth.py

@@ -1,377 +0,0 @@
-# type: ignore
-import argparse
-import json
-import os
-import statistics
-import sys
-import tempfile
-import time
-
-import mlx.core as mx
-
-DTYPE_MAP = {
-    "float32": (mx.float32, 4),
-    "float16": (mx.float16, 2),
-    "bfloat16": (mx.bfloat16, 2),
-}
-
-SIZES = [
-    1 * 1024,
-    4 * 1024,
-    16 * 1024,
-    64 * 1024,
-    256 * 1024,
-    1 * 1024 * 1024,
-    4 * 1024 * 1024,
-    16 * 1024 * 1024,
-    64 * 1024 * 1024,
-    256 * 1024 * 1024,
-    1 * 1024 * 1024 * 1024,
-    2 * 1024 * 1024 * 1024,
-    4 * 1024 * 1024 * 1024,
-    8 * 1024 * 1024 * 1024,
-]
-
-
-def format_bytes(n: int) -> str:
-    if n >= 1024 * 1024 * 1024:
-        return f"{n / (1024 * 1024 * 1024):.0f} GB"
-    if n >= 1024 * 1024:
-        return f"{n / (1024 * 1024):.0f} MB"
-    if n >= 1024:
-        return f"{n / 1024:.0f} KB"
-    return f"{n} B"
-
-
-def format_time(seconds: float) -> str:
-    if seconds >= 1.0:
-        return f"{seconds:.3f} s"
-    if seconds >= 0.001:
-        return f"{seconds * 1000:.2f} ms"
-    return f"{seconds * 1_000_000:.1f} us"
-
-
-def format_bandwidth(bytes_per_sec: float) -> str:
-    if bytes_per_sec >= 1024 * 1024 * 1024:
-        return f"{bytes_per_sec / (1024 * 1024 * 1024):.2f} GB/s"
-    if bytes_per_sec >= 1024 * 1024:
-        return f"{bytes_per_sec / (1024 * 1024):.1f} MB/s"
-    return f"{bytes_per_sec / 1024:.1f} KB/s"
-
-
-def barrier(group: mx.distributed.Group) -> None:
-    mx.eval(mx.distributed.all_sum(mx.array(1.0), group=group))
-
-
-def init_ring(
-    rank: int, self_ip: str, peer_ip: str, port: int, tmpdir: str
-) -> mx.distributed.Group:
-    if rank == 0:
-        hosts = [f"{self_ip}:{port}", f"{peer_ip}:{port}"]
-    else:
-        hosts = [f"{peer_ip}:{port}", f"{self_ip}:{port}"]
-
-    hostfile = os.path.join(tmpdir, "hosts.json")
-    with open(hostfile, "w") as f:
-        json.dump(hosts, f)
-
-    for var in ("MLX_HOSTFILE", "MLX_RANK", "MLX_IBV_DEVICES", "MLX_JACCL_COORDINATOR"):
-        os.environ.pop(var, None)
-
-    os.environ["MLX_HOSTFILE"] = hostfile
-    os.environ["MLX_RANK"] = str(rank)
-    return mx.distributed.init(backend="ring", strict=True)
-
-
-def init_jaccl(
-    rank: int, interface: str, coordinator: str, port: int, tmpdir: str
-) -> mx.distributed.Group:
-    devices = [[None, interface], [interface, None]]
-    devfile = os.path.join(tmpdir, "devices.json")
-    with open(devfile, "w") as f:
-        json.dump(devices, f)
-
-    for var in ("MLX_HOSTFILE", "MLX_RANK", "MLX_IBV_DEVICES", "MLX_JACCL_COORDINATOR"):
-        os.environ.pop(var, None)
-
-    os.environ["MLX_IBV_DEVICES"] = devfile
-    os.environ["MLX_RANK"] = str(rank)
-    if rank == 0:
-        os.environ["MLX_JACCL_COORDINATOR"] = f"0.0.0.0:{port}"
-    else:
-        os.environ["MLX_JACCL_COORDINATOR"] = coordinator
-
-    return mx.distributed.init(backend="jaccl", strict=True)
-
-
-def bench_unidirectional(
-    group: mx.distributed.Group,
-    rank: int,
-    size_bytes: int,
-    dtype: mx.Dtype,
-    element_size: int,
-    warmup: int,
-    iterations: int,
-) -> list[float]:
-    n_elements = size_bytes // element_size
-    tensor = mx.random.normal(shape=(n_elements,)).astype(dtype)
-    mx.eval(tensor)
-
-    for _ in range(warmup):
-        if rank == 0:
-            sent = mx.distributed.send(tensor, dst=1, group=group)
-            mx.eval(sent)
-        else:
-            received = mx.distributed.recv_like(tensor, src=0, group=group)
-            mx.eval(received)
-        barrier(group)
-
-    times: list[float] = []
-    for _ in range(iterations):
-        barrier(group)
-        t0 = time.perf_counter()
-        if rank == 0:
-            sent = mx.distributed.send(tensor, dst=1, group=group)
-            mx.eval(sent)
-        else:
-            received = mx.distributed.recv_like(tensor, src=0, group=group)
-            mx.eval(received)
-        barrier(group)
-        t1 = time.perf_counter()
-        times.append(t1 - t0)
-
-    return times
-
-
-def bench_rtt(
-    group: mx.distributed.Group,
-    rank: int,
-    size_bytes: int,
-    dtype: mx.Dtype,
-    element_size: int,
-    warmup: int,
-    iterations: int,
-) -> list[float]:
-    n_elements = size_bytes // element_size
-    tensor = mx.random.normal(shape=(n_elements,)).astype(dtype)
-    mx.eval(tensor)
-
-    for _ in range(warmup):
-        if rank == 0:
-            sent = mx.distributed.send(tensor, dst=1, group=group)
-            mx.eval(sent)
-            received = mx.distributed.recv_like(tensor, src=1, group=group)
-            mx.eval(received)
-        else:
-            received = mx.distributed.recv_like(tensor, src=0, group=group)
-            mx.eval(received)
-            sent = mx.distributed.send(received, dst=0, group=group)
-            mx.eval(sent)
-        barrier(group)
-
-    times: list[float] = []
-    for _ in range(iterations):
-        barrier(group)
-        t0 = time.perf_counter()
-        if rank == 0:
-            sent = mx.distributed.send(tensor, dst=1, group=group)
-            mx.eval(sent)
-            received = mx.distributed.recv_like(tensor, src=1, group=group)
-            mx.eval(received)
-        else:
-            received = mx.distributed.recv_like(tensor, src=0, group=group)
-            mx.eval(received)
-            sent = mx.distributed.send(received, dst=0, group=group)
-            mx.eval(sent)
-        barrier(group)
-        t1 = time.perf_counter()
-        times.append(t1 - t0)
-
-    return times
-
-
-def bench_all_gather(
-    group: mx.distributed.Group,
-    rank: int,
-    size_bytes: int,
-    dtype: mx.Dtype,
-    element_size: int,
-    warmup: int,
-    iterations: int,
-) -> list[float]:
-    n_elements = (size_bytes // 2) // element_size
-    tensor = mx.random.normal(shape=(n_elements,)).astype(dtype)
-    mx.eval(tensor)
-
-    for _ in range(warmup):
-        gathered = mx.distributed.all_gather(tensor, group=group)
-        mx.eval(gathered)
-        barrier(group)
-
-    times: list[float] = []
-    for _ in range(iterations):
-        barrier(group)
-        t0 = time.perf_counter()
-        gathered = mx.distributed.all_gather(tensor, group=group)
-        mx.eval(gathered)
-        t1 = time.perf_counter()
-        times.append(t1 - t0)
-
-    return times
-
-
-def print_table(title: str, rows: list[dict[str, str]]) -> None:
-    print(f"\n=== {title} ===")
-    headers = ["Size", "Median", "Min", "Max", "Bandwidth"]
-    widths = [
-        max(len(h), max((len(r[h]) for r in rows), default=0)) + 2 for h in headers
-    ]
-    header_line = "".join(h.ljust(w) for h, w in zip(headers, widths, strict=True))
-    print(header_line)
-    print("-" * len(header_line))
-    for row in rows:
-        print("".join(row[h].ljust(w) for h, w in zip(headers, widths, strict=True)))
-
-
-def run_bench(
-    name: str,
-    bench_fn,
-    group: mx.distributed.Group,
-    rank: int,
-    dtype: mx.Dtype,
-    element_size: int,
-    warmup: int,
-    iterations: int,
-    bw_multiplier: int = 1,
-) -> None:
-    rows: list[dict[str, str]] = []
-    for size in SIZES:
-        if rank == 0:
-            print(f"  {name}: {format_bytes(size)}...", end="", flush=True)
-        times = bench_fn(group, rank, size, dtype, element_size, warmup, iterations)
-        if rank == 0:
-            med = statistics.median(times)
-            mn = min(times)
-            mx_ = max(times)
-            bw = (size * bw_multiplier) / med
-            rows.append(
-                {
-                    "Size": format_bytes(size),
-                    "Median": format_time(med),
-                    "Min": format_time(mn),
-                    "Max": format_time(mx_),
-                    "Bandwidth": format_bandwidth(bw),
-                }
-            )
-            print(f" {format_bandwidth(bw)}")
-    if rank == 0:
-        print_table(name, rows)
-
-
-def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(
-        description="MLX Distributed Communication Benchmark"
-    )
-    subparsers = parser.add_subparsers(dest="backend", required=True)
-
-    ring_parser = subparsers.add_parser("ring")
-    ring_parser.add_argument("--rank", type=int, required=True, choices=[0, 1])
-    ring_parser.add_argument("--self-ip", required=True)
-    ring_parser.add_argument("--peer-ip", required=True)
-    ring_parser.add_argument("--port", type=int, default=5555)
-
-    jaccl_parser = subparsers.add_parser("jaccl")
-    jaccl_parser.add_argument("--rank", type=int, required=True, choices=[0, 1])
-    jaccl_parser.add_argument("--interface", required=True)
-    jaccl_parser.add_argument(
-        "--coordinator",
-        type=str,
-        default=None,
-        help="IP:PORT of rank 0 (required for rank 1)",
-    )
-    jaccl_parser.add_argument(
-        "--port", type=int, default=9999, help="Coordinator port (rank 0 only)"
-    )
-
-    for p in [ring_parser, jaccl_parser]:
-        p.add_argument("--warmup", type=int, default=3)
-        p.add_argument("--iterations", type=int, default=10)
-        p.add_argument("--dtype", choices=list(DTYPE_MAP.keys()), default="float32")
-
-    args = parser.parse_args()
-
-    if args.backend == "jaccl" and args.rank == 1 and args.coordinator is None:
-        jaccl_parser.error("--coordinator is required for rank 1")
-
-    return args
-
-
-def main() -> int:
-    args = parse_args()
-    dtype, element_size = DTYPE_MAP[args.dtype]
-
-    with tempfile.TemporaryDirectory() as tmpdir:
-        if args.backend == "ring":
-            print(f"Initializing ring backend (rank {args.rank})...")
-            group = init_ring(args.rank, args.self_ip, args.peer_ip, args.port, tmpdir)
-        else:
-            print(f"Initializing jaccl backend (rank {args.rank})...")
-            group = init_jaccl(
-                args.rank, args.interface, args.coordinator or "", args.port, tmpdir
-            )
-
-        print(f"Rank {group.rank()} of {group.size()} initialized")
-        barrier(group)
-
-        if args.rank == 0:
-            print("\nMLX Distributed Communication Benchmark")
-            print(
-                f"Backend: {args.backend} | Dtype: {args.dtype} | Warmup: {args.warmup} | Iterations: {args.iterations}"
-            )
-
-        run_bench(
-            "Unidirectional (rank 0 -> rank 1)",
-            bench_unidirectional,
-            group,
-            args.rank,
-            dtype,
-            element_size,
-            args.warmup,
-            args.iterations,
-        )
-        run_bench(
-            "Round-Trip (ping-pong)",
-            bench_rtt,
-            group,
-            args.rank,
-            dtype,
-            element_size,
-            args.warmup,
-            args.iterations,
-            bw_multiplier=2,
-        )
-        run_bench(
-            "All-Gather",
-            bench_all_gather,
-            group,
-            args.rank,
-            dtype,
-            element_size,
-            args.warmup,
-            args.iterations,
-        )
-
-        if args.rank == 0:
-            print("\nDone.")
-        else:
-            print("Rank 1 complete.")
-
-    return 0
-
-
-if __name__ == "__main__":
-    try:
-        sys.exit(main())
-    except KeyboardInterrupt:
-        print("\nInterrupted.")
-        sys.exit(1)

bench/vendor/lcb_testing_util.py

@@ -1,593 +0,0 @@
-# type: ignore
-# Vendored from LiveCodeBench (https://github.com/LiveCodeBench/LiveCodeBench)
-# File: lcb_runner/evaluation/testing_util.py
-# License: MIT
-# Vendored 2026-03-07 — do not modify without updating from upstream.
-
-import ast
-import faulthandler
-import json
-import platform
-
-# to run the solution files we're using a timing based approach
-import signal
-import sys
-import time
-
-# used for debugging to time steps
-from datetime import datetime
-from decimal import Decimal
-from enum import Enum
-from io import StringIO
-
-# from pyext import RuntimeModule
-from types import ModuleType
-
-# used for testing the code that reads from input
-from unittest.mock import mock_open, patch
-
-import_string = "from string import *\nfrom re import *\nfrom datetime import *\nfrom collections import *\nfrom heapq import *\nfrom bisect import *\nfrom copy import *\nfrom math import *\nfrom random import *\nfrom statistics import *\nfrom itertools import *\nfrom functools import *\nfrom operator import *\nfrom io import *\nfrom sys import *\nfrom json import *\nfrom builtins import *\nfrom typing import *\nimport string\nimport re\nimport datetime\nimport collections\nimport heapq\nimport bisect\nimport copy\nimport math\nimport random\nimport statistics\nimport itertools\nimport functools\nimport operator\nimport io\nimport sys\nimport json\nsys.setrecursionlimit(50000)\n"
-
-
-def truncatefn(s, length=300):
-    if isinstance(s, str):
-        pass
-    else:
-        s = str(s)
-    if len(s) <= length:
-        return s
-
-    return s[: length // 2] + "...(truncated) ..." + s[-length // 2 :]
-
-
-class CODE_TYPE(Enum):
-    call_based = 0
-    standard_input = 1
-
-
-# stuff for setting up signal timer
-class TimeoutException(Exception):
-    pass
-
-
-def timeout_handler(signum, frame):
-    print("timeout occured: alarm went off")
-    raise TimeoutException
-
-
-# used to capture stdout as a list
-# from https://stackoverflow.com/a/16571630/6416660
-# alternative use redirect_stdout() from contextlib
-class Capturing(list):
-    def __enter__(self):
-        self._stdout = sys.stdout
-        sys.stdout = self._stringio = StringIO()
-        # Make closing the StringIO a no-op
-        self._stringio.close = lambda x: 1
-        return self
-
-    def __exit__(self, *args):
-        self.append(self._stringio.getvalue())
-        del self._stringio  # free up some memory
-        sys.stdout = self._stdout
-
-
-# Custom mock for sys.stdin that supports buffer attribute
-class MockStdinWithBuffer:
-    def __init__(self, inputs: str):
-        self.inputs = inputs
-        self._stringio = StringIO(inputs)
-        self.buffer = MockBuffer(inputs)
-
-    def read(self, *args):
-        return self.inputs
-
-    def readline(self, *args):
-        return self._stringio.readline(*args)
-
-    def readlines(self, *args):
-        return self.inputs.split("\n")
-
-    def __getattr__(self, name):
-        # Delegate other attributes to StringIO
-        return getattr(self._stringio, name)
-
-
-class MockBuffer:
-    def __init__(self, inputs: str):
-        self.inputs = inputs.encode("utf-8")  # Convert to bytes
-
-    def read(self, *args):
-        # Return as byte strings that can be split
-        return self.inputs
-
-    def readline(self, *args):
-        return self.inputs.split(b"\n")[0] + b"\n"
-
-
-def clean_if_name(code: str) -> str:
-    try:
-        astree = ast.parse(code)
-        last_block = astree.body[-1]
-        if isinstance(last_block, ast.If):
-            condition = last_block.test
-            if ast.unparse(condition).strip() == "__name__ == '__main__'":
-                code = (
-                    ast.unparse(astree.body[:-1]) + "\n" + ast.unparse(last_block.body)  # type: ignore
-                )
-    except:
-        pass
-
-    return code
-
-
-def make_function(code: str) -> str:
-    try:
-        import_stmts = []
-        all_other_stmts = []
-        astree = ast.parse(code)
-        for stmt in astree.body:
-            if isinstance(stmt, (ast.Import, ast.ImportFrom)):
-                import_stmts.append(stmt)
-            else:
-                all_other_stmts.append(stmt)
-
-        function_ast = ast.FunctionDef(
-            name="wrapped_function",
-            args=ast.arguments(
-                posonlyargs=[], args=[], kwonlyargs=[], kw_defaults=[], defaults=[]
-            ),
-            body=all_other_stmts,
-            decorator_list=[],
-            lineno=-1,
-        )
-        main_code = (
-            import_string
-            + "\n"
-            + ast.unparse(import_stmts)
-            + "\n"
-            + ast.unparse(function_ast)
-        )
-        return main_code
-    except Exception:
-        return code
-
-
-def call_method(method, inputs):
-    if isinstance(inputs, list):
-        inputs = "\n".join(inputs)
-
-    inputs_line_iterator = iter(inputs.split("\n"))
-
-    # Create custom stdin mock with buffer support
-    mock_stdin = MockStdinWithBuffer(inputs)
-
-    # sys.setrecursionlimit(10000)
-
-    # @patch('builtins.input', side_effect=inputs.split("\n"))
-    @patch("builtins.open", mock_open(read_data=inputs))
-    @patch("sys.stdin", mock_stdin)  # Use our custom mock instead of StringIO
-    @patch("sys.stdin.readline", lambda *args: next(inputs_line_iterator))
-    @patch("sys.stdin.readlines", lambda *args: inputs.split("\n"))
-    @patch("sys.stdin.read", lambda *args: inputs)
-    # @patch('sys.stdout.write', print)
-    def _inner_call_method(_method):
-        try:
-            return _method()
-        except SystemExit:
-            pass
-        finally:
-            pass
-
-    return _inner_call_method(method)
-
-
-def get_function(compiled_sol, fn_name: str):  # type: ignore
-    try:
-        assert hasattr(compiled_sol, fn_name)
-        return getattr(compiled_sol, fn_name)
-    except Exception:
-        return
-
-
-def compile_code(code: str, timeout: int):
-    signal.alarm(timeout)
-    try:
-        tmp_sol = ModuleType("tmp_sol", "")
-        exec(code, tmp_sol.__dict__)
-        if "class Solution" in code:
-            # leetcode wraps solutions in `Solution`
-            # this is a hack to check if it is leetcode solution or not
-            # currently livecodebench only supports LeetCode but
-            # else condition allows future extensibility to other platforms
-            compiled_sol = tmp_sol.Solution()
-        else:
-            # do nothing in the other case since function is accesible
-            compiled_sol = tmp_sol
-
-        assert compiled_sol is not None
-    finally:
-        signal.alarm(0)
-
-    return compiled_sol
-
-
-def convert_line_to_decimals(line: str) -> tuple[bool, list[Decimal]]:
-    try:
-        decimal_line = [Decimal(elem) for elem in line.split()]
-    except:
-        return False, []
-    return True, decimal_line
-
-
-def get_stripped_lines(val: str):
-    ## you don't want empty lines to add empty list after splitlines!
-    val = val.strip()
-
-    return [val_line.strip() for val_line in val.split("\n")]
-
-
-def grade_call_based(
-    code: str, all_inputs: list, all_outputs: list, fn_name: str, timeout: int
-):
-    # call-based clean up logic
-    # need to wrap in try-catch logic after to catch the correct errors, but for now this is fine.
-    code = import_string + "\n\n" + code
-    compiled_sol = compile_code(code, timeout)
-
-    if compiled_sol is None:
-        return
-
-    method = get_function(compiled_sol, fn_name)
-
-    if method is None:
-        return
-
-    all_inputs = [
-        [json.loads(line) for line in inputs.split("\n")] for inputs in all_inputs
-    ]
-
-    all_outputs = [json.loads(output) for output in all_outputs]
-
-    total_execution = 0
-    all_results = []
-    for idx, (gt_inp, gt_out) in enumerate(zip(all_inputs, all_outputs)):
-        signal.alarm(timeout)
-        faulthandler.enable()
-        try:
-            # can lock here so time is useful
-            start = time.time()
-            prediction = method(*gt_inp)
-            total_execution += time.time() - start
-            signal.alarm(0)
-
-            # don't penalize model if it produces tuples instead of lists
-            # ground truth sequences are not tuples
-            if isinstance(prediction, tuple):
-                prediction = list(prediction)
-
-            tmp_result = prediction == gt_out
-
-            # handle floating point comparisons
-
-            all_results.append(tmp_result)
-
-            if not tmp_result:
-                return all_results, {
-                    "output": truncatefn(prediction),
-                    "inputs": truncatefn(gt_inp),
-                    "expected": truncatefn(gt_out),
-                    "error_code": -2,
-                    "error_message": "Wrong Answer",
-                }
-        except Exception as e:
-            signal.alarm(0)
-            if "timeoutexception" in repr(e).lower():
-                all_results.append(-3)
-                return all_results, {
-                    "error": repr(e),
-                    "error_code": -3,
-                    "error_message": "Time Limit Exceeded",
-                    "inputs": truncatefn(gt_inp),
-                    "expected": truncatefn(gt_out),
-                }
-            else:
-                all_results.append(-4)
-                return all_results, {
-                    "error": repr(e),
-                    "error_code": -4,
-                    "error_message": "Runtime Error",
-                    "inputs": truncatefn(gt_inp),
-                    "expected": truncatefn(gt_out),
-                }
-
-        finally:
-            signal.alarm(0)
-            faulthandler.disable()
-
-    return all_results, {"execution time": total_execution}
-
-
-def grade_stdio(
-    code: str,
-    all_inputs: list,
-    all_outputs: list,
-    timeout: int,
-):
-    ## runtime doesn't interact well with __name__ == '__main__'
-    code = clean_if_name(code)
-
-    ## we wrap the given code inside another function
-    code = make_function(code)
-
-    compiled_sol = compile_code(code, timeout)
-    if compiled_sol is None:
-        return
-
-    method = get_function(compiled_sol, "wrapped_function")
-
-    if method is None:
-        return
-
-    all_results = []
-    total_execution_time = 0
-    for idx, (gt_inp, gt_out) in enumerate(zip(all_inputs, all_outputs)):
-        signal.alarm(timeout)
-        faulthandler.enable()
-
-        signal.alarm(timeout)
-        with Capturing() as captured_output:
-            try:
-                start = time.time()
-                call_method(method, gt_inp)
-                total_execution_time += time.time() - start
-                # reset the alarm
-                signal.alarm(0)
-            except Exception as e:
-                signal.alarm(0)
-                if "timeoutexception" in repr(e).lower():
-                    all_results.append(-3)
-                    return all_results, {
-                        "error": repr(e),
-                        "error_code": -3,
-                        "error_message": "Time Limit Exceeded",
-                        "inputs": truncatefn(gt_inp),
-                        "expected": truncatefn(gt_out),
-                    }
-                else:
-                    all_results.append(-4)
-                    return all_results, {
-                        "error": repr(e),
-                        "error_code": -4,
-                        "error_message": "Runtime Error",
-                        "inputs": truncatefn(gt_inp),
-                        "expected": truncatefn(gt_out),
-                    }
-
-            finally:
-                signal.alarm(0)
-                faulthandler.disable()
-
-        prediction = captured_output[0]
-
-        stripped_prediction_lines = get_stripped_lines(prediction)
-        stripped_gt_out_lines = get_stripped_lines(gt_out)
-
-        ## WA happens in multiple circumstances
-        ## so cache the return to make it clean!
-        WA_send_args = {
-            "output": truncatefn(prediction),
-            "inputs": truncatefn(gt_inp),
-            "expected": truncatefn(gt_out),
-            "error_code": -2,
-        }
-
-        if len(stripped_prediction_lines) != len(stripped_gt_out_lines):
-            all_results.append(-2)
-            WA_send_args["error_message"] = "Wrong answer: mismatched output length"
-            return all_results, WA_send_args
-
-        for output_line_idx, (
-            stripped_prediction_line,
-            stripped_gt_out_line,
-        ) in enumerate(zip(stripped_prediction_lines, stripped_gt_out_lines)):
-            WA_send_args["error_message"] = (
-                f"Wrong answer at {output_line_idx=}: {truncatefn(stripped_prediction_line)} != {truncatefn(stripped_gt_out_line)}"
-            )
-
-            ## CASE 1: exact match
-            if stripped_prediction_line == stripped_gt_out_line:
-                continue
-
-            ## CASE 2: element-wise comparision
-            ## if there are floating elements
-            ## use `decimal` library for good floating point comparision
-            ## otherwise gotcha: np.isclose(50000000000000000, 50000000000000001) = True
-            ## note that we should always be able to convert to decimals
-
-            success, decimal_prediction_line = convert_line_to_decimals(
-                stripped_prediction_line
-            )
-            if not success:
-                all_results.append(-2)
-                return all_results, WA_send_args
-            success, decimal_gtout_line = convert_line_to_decimals(stripped_gt_out_line)
-            if not success:
-                all_results.append(-2)
-                return all_results, WA_send_args
-
-            if decimal_prediction_line == decimal_gtout_line:
-                continue
-
-            all_results.append(-2)
-            return all_results, WA_send_args
-        all_results.append(True)
-
-    return all_results, {"execution time": total_execution_time}
-
-
-def run_test(sample, test=None, debug=False, timeout=6):
-    """
-    if test(generated_code) is not None it'll try to run the code.
-    otherwise it'll just return an input and output pair.
-    """
-    signal.signal(signal.SIGALRM, timeout_handler)
-
-    # Disable functionalities that can make destructive changes to the test.
-    # max memory is set to 4GB
-    reliability_guard()
-
-    if debug:
-        print(f"start = {datetime.now().time()}")
-
-    try:
-        in_outs = json.loads(sample["input_output"])
-    except ValueError as e:
-        raise e
-        in_outs = None
-
-    if in_outs:
-        if in_outs.get("fn_name") is None:
-            which_type = CODE_TYPE.standard_input  # Standard input
-            method_name = None
-
-        else:
-            which_type = CODE_TYPE.call_based  # Call-based
-            method_name = in_outs["fn_name"]
-
-    if debug:
-        print(f"loaded input_output = {datetime.now().time()}")
-
-    if test is None:
-        assert False, "should not happen: test code is none"
-        return in_outs, {"error": "no test code provided"}
-    elif test is not None:
-        results = []
-        sol = import_string
-        if debug:
-            print(f"loading test code = {datetime.now().time()}")
-
-        if which_type == CODE_TYPE.call_based:
-            signal.alarm(timeout)
-            try:
-                results, metadata = grade_call_based(
-                    code=test,
-                    all_inputs=in_outs["inputs"],
-                    all_outputs=in_outs["outputs"],
-                    fn_name=method_name,
-                    timeout=timeout,
-                )
-                return results, metadata
-            except Exception as e:
-                return [-4], {
-                    "error_code": -4,
-                    "error_message": f"Error during testing: {e}",
-                }
-            finally:
-                signal.alarm(0)
-        elif which_type == CODE_TYPE.standard_input:
-            # sol
-            # if code has if __name__ == "__main__": then remove it
-
-            signal.alarm(timeout)
-            try:
-                results, metadata = grade_stdio(
-                    code=test,
-                    all_inputs=in_outs["inputs"],
-                    all_outputs=in_outs["outputs"],
-                    timeout=timeout,
-                )
-                return results, metadata
-            except Exception as e:
-                return [-4], {
-                    "error_code": -4,
-                    "error_message": f"Error during testing: {e}",
-                }
-            finally:
-                signal.alarm(0)
-
-
-def reliability_guard(maximum_memory_bytes=None):
-    """
-    This disables various destructive functions and prevents the generated code
-    from interfering with the test (e.g. fork bomb, killing other processes,
-    removing filesystem files, etc.)
-    WARNING
-    This function is NOT a security sandbox. Untrusted code, including, model-
-    generated code, should not be blindly executed outside of one. See the
-    Codex paper for more information about OpenAI's code sandbox, and proceed
-    with caution.
-    """
-
-    if maximum_memory_bytes is not None:
-        import resource
-
-        resource.setrlimit(
-            resource.RLIMIT_AS, (maximum_memory_bytes, maximum_memory_bytes)
-        )
-        resource.setrlimit(
-            resource.RLIMIT_DATA, (maximum_memory_bytes, maximum_memory_bytes)
-        )
-        if not platform.uname().system == "Darwin":
-            resource.setrlimit(
-                resource.RLIMIT_STACK, (maximum_memory_bytes, maximum_memory_bytes)
-            )
-
-    faulthandler.disable()
-
-    import builtins
-
-    # builtins.exit = None
-    builtins.quit = None
-
-    import os
-
-    os.environ["OMP_NUM_THREADS"] = "1"
-
-    os.kill = None
-    os.system = None
-    os.putenv = None
-    os.remove = None
-    os.removedirs = None
-    os.rmdir = None
-    os.fchdir = None
-    os.setuid = None
-    os.fork = None
-    os.forkpty = None
-    os.killpg = None
-    os.rename = None
-    os.renames = None
-    os.truncate = None
-    os.replace = None
-    os.unlink = None
-    os.fchmod = None
-    os.fchown = None
-    os.chmod = None
-    os.chown = None
-    os.chroot = None
-    os.fchdir = None
-    os.lchflags = None
-    os.lchmod = None
-    os.lchown = None
-    os.getcwd = None
-    os.chdir = None
-
-    import shutil
-
-    shutil.rmtree = None
-    shutil.move = None
-    shutil.chown = None
-
-    import subprocess
-
-    subprocess.Popen = None
-
-    __builtins__["help"] = None
-
-    import sys
-
-    sys.modules["ipdb"] = None
-    sys.modules["joblib"] = None
-    sys.modules["resource"] = None
-    sys.modules["psutil"] = None
-    sys.modules["tkinter"] = None

dashboard/src/lib/components/ChatForm.svelte

@@ -1,6 +1,9 @@
 <script lang="ts">
   import {
     isLoading,
+    sendMessage,
+    generateImage,
+    editImage,
     editingImage,
     clearEditingImage,
     selectedChatModel,
@@ -25,7 +28,7 @@
     modelTasks?: Record<string, string[]>;
     modelCapabilities?: Record<string, string[]>;
     onSend?: () => void;
-    onAutoSend: (
+    onAutoSend?: (
       content: string,
       files?: {
         id: string;
@@ -213,10 +216,49 @@
     uploadedFiles = [];
     resetTextareaHeight();
 
-    // Parent controls all send logic (including image routing,
-    // launching non-running models before sending, etc.)
-    onAutoSend(content, files);
+    // When onAutoSend is provided, the parent controls all send logic
+    // (including launching non-running models before sending)
+    if (onAutoSend) {
+      onAutoSend(content, files);
+      onSend?.();
+      setTimeout(() => textareaRef?.focus(), 10);
+      return;
+    }
+
+    // Use image editing if in edit mode
+    if (isEditMode && currentEditingImage && content) {
+      editImage(content, currentEditingImage.imageDataUrl);
+    }
+    // If user attached an image with an ImageToImage model, use edit endpoint
+    else if (
+      currentModel &&
+      modelSupportsImageEditing(currentModel) &&
+      files.length > 0 &&
+      content
+    ) {
+      // Use the first attached image for editing
+      const imageFile = files[0];
+      if (imageFile.preview) {
+        editImage(content, imageFile.preview);
+      }
+    } else if (
+      currentModel &&
+      modelSupportsTextToImage(currentModel) &&
+      content
+    ) {
+      // Use image generation for text-to-image models
+      generateImage(content);
+    } else {
+      sendMessage(
+        content,
+        files,
+        modelSupportsThinking() ? thinkingEnabled : null,
+      );
+    }
+
     onSend?.();
+
+    // Refocus the textarea after sending
     setTimeout(() => textareaRef?.focus(), 10);
   }
 

dashboard/src/lib/components/ChatSidebar.svelte

@@ -18,18 +18,12 @@
     class?: string;
     onNewChat?: () => void;
     onSelectConversation?: () => void;
-    isMobileDrawer?: boolean;
-    isOpen?: boolean;
-    onClose?: () => void;
   }
 
   let {
     class: className = "",
     onNewChat,
     onSelectConversation,
-    isMobileDrawer = false,
-    isOpen = false,
-    onClose,
   }: Props = $props();
 
   const conversationList = $derived(conversations());
@@ -59,10 +53,6 @@
   function handleSelectConversation(id: string) {
     onSelectConversation?.();
     loadConversation(id);
-    // Close mobile drawer when selecting a conversation
-    if (isMobileDrawer && isOpen) {
-      onClose?.();
-    }
   }
 
   function handleStartEdit(id: string, name: string, event: MouseEvent) {
@@ -262,7 +252,9 @@
   }
 </script>
 
-{#snippet sidebarContent()}
+<aside
+  class="flex flex-col h-full bg-exo-dark-gray border-r border-exo-yellow/10 {className}"
+>
   <!-- Header -->
   <div class="p-4">
     <button
@@ -599,30 +591,4 @@
       </button>
     </div>
   </div>
-{/snippet}
-
-{#if isMobileDrawer}
-  <!-- Mobile drawer with overlay -->
-  {#if isOpen}
-    <!-- Overlay backdrop -->
-    <button
-      type="button"
-      class="fixed inset-0 bg-black/60 backdrop-blur-sm z-40 md:hidden"
-      onclick={() => onClose?.()}
-      aria-label="Close sidebar"
-    ></button>
-    <!-- Drawer panel -->
-    <aside
-      class="fixed left-0 top-0 bottom-0 w-72 bg-exo-dark-gray border-r border-exo-yellow/10 z-50 flex flex-col md:hidden"
-    >
-      {@render sidebarContent()}
-    </aside>
-  {/if}
-{:else}
-  <!-- Desktop sidebar -->
-  <aside
-    class="flex flex-col h-full bg-exo-dark-gray border-r border-exo-yellow/10 {className}"
-  >
-    {@render sidebarContent()}
-  </aside>
-{/if}
+</aside>

dashboard/src/lib/components/FamilyLogos.svelte

@@ -82,12 +82,6 @@
       d="M12.025 1.13c-5.77 0-10.449 4.647-10.449 10.378 0 1.112.178 2.181.503 3.185.064-.222.203-.444.416-.577a.96.96 0 0 1 .524-.15c.293 0 .584.124.84.284.278.173.48.408.71.694.226.282.458.611.684.951v-.014c.017-.324.106-.622.264-.874s.403-.487.762-.543c.3-.047.596.06.787.203s.31.313.4.467c.15.257.212.468.233.542.01.026.653 1.552 1.657 2.54.616.605 1.01 1.223 1.082 1.912.055.537-.096 1.059-.38 1.572.637.121 1.294.187 1.967.187.657 0 1.298-.063 1.921-.178-.287-.517-.44-1.041-.384-1.581.07-.69.465-1.307 1.081-1.913 1.004-.987 1.647-2.513 1.657-2.539.021-.074.083-.285.233-.542.09-.154.208-.323.4-.467a1.08 1.08 0 0 1 .787-.203c.359.056.604.29.762.543s.247.55.265.874v.015c.225-.34.457-.67.683-.952.23-.286.432-.52.71-.694.257-.16.547-.284.84-.285a.97.97 0 0 1 .524.151c.228.143.373.388.43.625l.006.04a10.3 10.3 0 0 0 .534-3.273c0-5.731-4.678-10.378-10.449-10.378M8.327 6.583a1.5 1.5 0 0 1 .713.174 1.487 1.487 0 0 1 .617 2.013c-.183.343-.762-.214-1.102-.094-.38.134-.532.914-.917.71a1.487 1.487 0 0 1 .69-2.803m7.486 0a1.487 1.487 0 0 1 .689 2.803c-.385.204-.536-.576-.916-.71-.34-.12-.92.437-1.103.094a1.487 1.487 0 0 1 .617-2.013 1.5 1.5 0 0 1 .713-.174m-10.68 1.55a.96.96 0 1 1 0 1.921.96.96 0 0 1 0-1.92m13.838 0a.96.96 0 1 1 0 1.92.96.96 0 0 1 0-1.92M8.489 11.458c.588.01 1.965 1.157 3.572 1.164 1.607-.007 2.984-1.155 3.572-1.164.196-.003.305.12.305.454 0 .886-.424 2.328-1.563 3.202-.22-.756-1.396-1.366-1.63-1.32q-.011.001-.02.006l-.044.026-.01.008-.03.024q-.018.017-.035.036l-.032.04a1 1 0 0 0-.058.09l-.014.025q-.049.088-.11.19a1 1 0 0 1-.083.116 1.2 1.2 0 0 1-.173.18q-.035.029-.075.058a1.3 1.3 0 0 1-.251-.243 1 1 0 0 1-.076-.107c-.124-.193-.177-.363-.337-.444-.034-.016-.104-.008-.2.022q-.094.03-.216.087-.06.028-.125.063l-.13.074q-.067.04-.136.086a3 3 0 0 0-.135.096 3 3 0 0 0-.26.219 2 2 0 0 0-.12.121 2 2 0 0 0-.106.128l-.002.002a2 2 0 0 0-.09.132l-.001.001a1.2 1.2 0 0 0-.105.212q-.013.036-.024.073c-1.139-.875-1.563-2.317-1.563-3.203 0-.334.109-.457.305-.454m.836 10.354c.824-1.19.766-2.082-.365-3.194-1.13-1.112-1.789-2.738-1.789-2.738s-.246-.945-.806-.858-.97 1.499.202 2.362c1.173.864-.233 1.45-.685.64-.45-.812-1.683-2.896-2.322-3.295s-1.089-.175-.938.647 2.822 2.813 2.562 3.244-1.176-.506-1.176-.506-2.866-2.567-3.49-1.898.473 1.23 2.037 2.16c1.564.932 1.686 1.178 1.464 1.53s-3.675-2.511-4-1.297c-.323 1.214 3.524 1.567 3.287 2.405-.238.839-2.71-1.587-3.216-.642-.506.946 3.49 2.056 3.522 2.064 1.29.33 4.568 1.028 5.713-.624m5.349 0c-.824-1.19-.766-2.082.365-3.194 1.13-1.112 1.789-2.738 1.789-2.738s.246-.945.806-.858.97 1.499-.202 2.362c-1.173.864.233 1.45.685.64.451-.812 1.683-2.896 2.322-3.295s1.089-.175.938.647-2.822 2.813-2.562 3.244 1.176-.506 1.176-.506 2.866-2.567 3.49-1.898-.473 1.23-2.037 2.16c-1.564.932-1.686 1.178-1.464 1.53s3.675-2.511 4-1.297c.323 1.214-3.524 1.567-3.287 2.405.238.839 2.71-1.587 3.216-.642.506.946-3.49 2.056-3.522 2.064-1.29.33-4.568 1.028-5.713-.624"
     />
   </svg>
-{:else if family === "step"}
-  <svg class="w-6 h-6 {className}" viewBox="0 0 24 24" fill="currentColor">
-    <path
-      d="M22.012 0h1.032v.927H24v.968h-.956V3.78h-1.032V1.896h-1.878v-.97h1.878V0zM2.6 12.371V1.87h.969v10.502h-.97zm10.423.66h10.95v.918h-6.208v9.579h-4.742V13.03zM5.629 3.333v12.356H0v4.51h10.386V8L20.859 8l-.003-4.668-15.227.001z"
-    />
-  </svg>
 {:else}
   <svg class="w-6 h-6 {className}" viewBox="0 0 24 24" fill="currentColor">
     <path

dashboard/src/lib/components/FamilySidebar.svelte

@@ -41,13 +41,13 @@
 </script>
 
 <div
-  class="flex flex-col gap-1 py-2 px-1 border-r border-exo-yellow/10 bg-exo-medium-gray/30 min-w-[72px] sm:min-w-[64px] overflow-y-auto scrollbar-hide"
+  class="flex flex-col gap-1 py-2 px-1 border-r border-exo-yellow/10 bg-exo-medium-gray/30 min-w-[64px] overflow-y-auto scrollbar-hide"
 >
   <!-- All models (no filter) -->
   <button
     type="button"
     onclick={() => onSelect(null)}
-    class="group flex flex-col items-center justify-center p-2 sm:p-2 rounded transition-all duration-200 cursor-pointer min-h-[44px] sm:min-h-0 {selectedFamily ===
+    class="group flex flex-col items-center justify-center p-2 rounded transition-all duration-200 cursor-pointer {selectedFamily ===
     null
       ? 'bg-exo-yellow/20 border-l-2 border-exo-yellow'
       : 'hover:bg-white/5 border-l-2 border-transparent'}"

dashboard/src/lib/components/HeaderNav.svelte

@@ -6,12 +6,6 @@
   export let showSidebarToggle = false;
   export let sidebarVisible = true;
   export let onToggleSidebar: (() => void) | null = null;
-  export let showMobileMenuToggle = false;
-  export let mobileMenuOpen = false;
-  export let onToggleMobileMenu: (() => void) | null = null;
-  export let showMobileRightToggle = false;
-  export let mobileRightOpen = false;
-  export let onToggleMobileRight: (() => void) | null = null;
   export let downloadProgress: {
     count: number;
     percentage: number;
@@ -33,96 +27,49 @@
       onToggleSidebar();
     }
   }
-
-  function handleToggleMobileMenu(): void {
-    if (onToggleMobileMenu) {
-      onToggleMobileMenu();
-    }
-  }
-
-  function handleToggleMobileRight(): void {
-    if (onToggleMobileRight) {
-      onToggleMobileRight();
-    }
-  }
 </script>
 
 <header
-  class="relative z-20 flex items-center justify-center px-4 md:px-6 pt-4 md:pt-8 pb-3 md:pb-4 bg-exo-dark-gray"
+  class="relative z-20 flex items-center justify-center px-6 pt-8 pb-4 bg-exo-dark-gray"
 >
-  <!-- Left: Sidebar Toggle (desktop) or Mobile Sidebar Toggle (mobile) -->
-  <div
-    class="absolute left-4 md:left-6 top-1/2 -translate-y-1/2 flex items-center gap-2"
-  >
-    <!-- Mobile sidebar toggle -->
-    <button
-      onclick={handleToggleMobileMenu}
-      class="p-2 rounded border border-exo-light-gray/30 hover:border-exo-yellow/50 hover:bg-exo-medium-gray/30 transition-colors cursor-pointer md:hidden"
-      title={mobileMenuOpen ? "Hide sidebar" : "Show sidebar"}
-      aria-label={mobileMenuOpen
-        ? "Hide conversation sidebar"
-        : "Show conversation sidebar"}
-      aria-pressed={mobileMenuOpen}
-    >
-      <svg
-        fill="none"
-        viewBox="0 0 24 24"
-        stroke="currentColor"
-        stroke-width="2"
-        class="w-5 h-5 {mobileMenuOpen
-          ? 'text-exo-yellow'
-          : 'text-exo-light-gray'}"
+  <!-- Left: Sidebar Toggle -->
+  {#if showSidebarToggle}
+    <div class="absolute left-6 top-1/2 -translate-y-1/2">
+      <button
+        onclick={handleToggleSidebar}
+        class="p-2 rounded border border-exo-light-gray/30 hover:border-exo-yellow/50 hover:bg-exo-medium-gray/30 transition-colors cursor-pointer"
+        title={sidebarVisible ? "Hide sidebar" : "Show sidebar"}
+        aria-label={sidebarVisible
+          ? "Hide conversation sidebar"
+          : "Show conversation sidebar"}
+        aria-pressed={sidebarVisible}
       >
-        {#if mobileMenuOpen}
-          <path
-            stroke-linecap="round"
-            stroke-linejoin="round"
-            d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
-          ></path>
-        {:else}
-          <path
-            stroke-linecap="round"
-            stroke-linejoin="round"
-            d="M13 5l7 7-7 7M5 5l7 7-7 7"
-          ></path>
-        {/if}
-      </svg>
-    </button>
-    <!-- Desktop sidebar toggle -->
-    <button
-      onclick={handleToggleSidebar}
-      class="p-2 rounded border border-exo-light-gray/30 hover:border-exo-yellow/50 hover:bg-exo-medium-gray/30 transition-colors cursor-pointer hidden md:block"
-      title={sidebarVisible ? "Hide sidebar" : "Show sidebar"}
-      aria-label={sidebarVisible
-        ? "Hide conversation sidebar"
-        : "Show conversation sidebar"}
-      aria-pressed={sidebarVisible}
-    >
-      <svg
-        fill="none"
-        viewBox="0 0 24 24"
-        stroke="currentColor"
-        stroke-width="2"
-        class="w-5 h-5 {sidebarVisible
-          ? 'text-exo-yellow'
-          : 'text-exo-light-gray'}"
-      >
-        {#if sidebarVisible}
-          <path
-            stroke-linecap="round"
-            stroke-linejoin="round"
-            d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
-          ></path>
-        {:else}
-          <path
-            stroke-linecap="round"
-            stroke-linejoin="round"
-            d="M13 5l7 7-7 7M5 5l7 7-7 7"
-          ></path>
-        {/if}
-      </svg>
-    </button>
-  </div>
+        <svg
+          class="w-5 h-5 {sidebarVisible
+            ? 'text-exo-yellow'
+            : 'text-exo-light-gray'}"
+          fill="none"
+          viewBox="0 0 24 24"
+          stroke="currentColor"
+          stroke-width="2"
+        >
+          {#if sidebarVisible}
+            <path
+              stroke-linecap="round"
+              stroke-linejoin="round"
+              d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
+            />
+          {:else}
+            <path
+              stroke-linecap="round"
+              stroke-linejoin="round"
+              d="M13 5l7 7-7 7M5 5l7 7-7 7"
+            />
+          {/if}
+        </svg>
+      </button>
+    </div>
+  {/if}
 
   <!-- Center: Logo (clickable to go home) -->
   <button
@@ -136,55 +83,19 @@
     <img
       src="/exo-logo.png"
       alt="EXO"
-      class="h-12 md:h-18 drop-shadow-[0_0_4px_rgba(255,215,0,0.3)]"
+      class="h-18 drop-shadow-[0_0_4px_rgba(255,215,0,0.3)]"
     />
   </button>
 
-  <!-- Right: Home + Downloads + Mobile Right Toggle -->
+  <!-- Right: Home + Downloads -->
   <nav
-    class="absolute right-4 md:right-6 top-1/2 -translate-y-1/2 flex items-center gap-2 md:gap-4"
+    class="absolute right-6 top-1/2 -translate-y-1/2 flex items-center gap-4"
     aria-label="Main navigation"
   >
-    <!-- Mobile right sidebar toggle (instances/models) - only show when not in chat mode -->
-    {#if showMobileRightToggle}
-      <button
-        onclick={handleToggleMobileRight}
-        class="p-2 rounded border border-exo-light-gray/30 hover:border-exo-yellow/50 hover:bg-exo-medium-gray/30 transition-colors cursor-pointer md:hidden"
-        title={mobileRightOpen ? "Hide instances" : "Show instances"}
-        aria-label={mobileRightOpen
-          ? "Hide instances panel"
-          : "Show instances panel"}
-        aria-pressed={mobileRightOpen}
-      >
-        <svg
-          fill="none"
-          viewBox="0 0 24 24"
-          stroke="currentColor"
-          stroke-width="2"
-          class="w-5 h-5 {mobileRightOpen
-            ? 'text-exo-yellow'
-            : 'text-exo-light-gray'}"
-        >
-          {#if mobileRightOpen}
-            <path
-              stroke-linecap="round"
-              stroke-linejoin="round"
-              d="M13 5l7 7-7 7M5 5l7 7-7 7"
-            ></path>
-          {:else}
-            <path
-              stroke-linecap="round"
-              stroke-linejoin="round"
-              d="M11 19l-7-7 7-7m8 14l-7-7 7-7"
-            ></path>
-          {/if}
-        </svg>
-      </button>
-    {/if}
     {#if showHome}
       <button
         onclick={handleHome}
-        class="flex text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase items-center gap-2 cursor-pointer"
+        class="text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
         title="Back to topology view"
       >
         <svg
@@ -200,12 +111,12 @@
             d="M3 12l2-2m0 0l7-7 7 7M5 10v10a1 1 0 001 1h3m10-11l2 2m-2-2v10a1 1 0 01-1 1h-3m-6 0a1 1 0 001-1v-4a1 1 0 011-1h2a1 1 0 011 1v4a1 1 0 001 1m-6 0h6"
           />
         </svg>
-        <span class="hidden sm:inline">Home</span>
+        Home
       </button>
     {/if}
     <a
       href="/#/downloads"
-      class="text-xs md:text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-1.5 md:gap-2 cursor-pointer"
+      class="text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
       title="View downloads overview"
     >
       {#if downloadProgress}
@@ -257,7 +168,7 @@
           <path d="M5 21h14" />
         </svg>
       {/if}
-      <span class="hidden sm:inline">Downloads</span>
+      Downloads
     </a>
   </nav>
 </header>

dashboard/src/lib/stores/app.svelte.ts

@@ -580,8 +580,6 @@ class AppStore {
   debugMode = $state(false);
   topologyOnlyMode = $state(false);
   chatSidebarVisible = $state(true); // Shown by default
-  mobileChatSidebarOpen = $state(false); // Mobile drawer state
-  mobileRightSidebarOpen = $state(false); // Mobile right drawer state
 
   // Image generation params
   imageGenerationParams = $state<ImageGenerationParams>({
@@ -1247,30 +1245,6 @@ class AppStore {
     this.saveChatSidebarVisibleToStorage();
   }
 
-  getMobileChatSidebarOpen(): boolean {
-    return this.mobileChatSidebarOpen;
-  }
-
-  setMobileChatSidebarOpen(open: boolean) {
-    this.mobileChatSidebarOpen = open;
-  }
-
-  toggleMobileChatSidebar() {
-    this.mobileChatSidebarOpen = !this.mobileChatSidebarOpen;
-  }
-
-  getMobileRightSidebarOpen(): boolean {
-    return this.mobileRightSidebarOpen;
-  }
-
-  setMobileRightSidebarOpen(open: boolean) {
-    this.mobileRightSidebarOpen = open;
-  }
-
-  toggleMobileRightSidebar() {
-    this.mobileRightSidebarOpen = !this.mobileRightSidebarOpen;
-  }
-
   startPolling() {
     this.fetchState();
     this.fetchInterval = setInterval(() => this.fetchState(), 1000);
@@ -1577,7 +1551,6 @@ class AppStore {
     // Remove messages after user message (including the user message for image requests
     // since generateImage/editImage will re-add it)
     this.messages = this.messages.slice(0, lastUserIndex);
-    this.updateActiveConversation();
 
     switch (requestType) {
       case "image-generation":
@@ -3309,18 +3282,6 @@ export const toggleChatSidebarVisible = () =>
   appStore.toggleChatSidebarVisible();
 export const setChatSidebarVisible = (visible: boolean) =>
   appStore.setChatSidebarVisible(visible);
-
-// Mobile sidebar state
-export const mobileChatSidebarOpen = () => appStore.mobileChatSidebarOpen;
-export const toggleMobileChatSidebar = () => appStore.toggleMobileChatSidebar();
-export const setMobileChatSidebarOpen = (open: boolean) =>
-  appStore.setMobileChatSidebarOpen(open);
-export const mobileRightSidebarOpen = () => appStore.mobileRightSidebarOpen;
-export const toggleMobileRightSidebar = () =>
-  appStore.toggleMobileRightSidebar();
-export const setMobileRightSidebarOpen = (open: boolean) =>
-  appStore.setMobileRightSidebarOpen(open);
-
 export const refreshState = () => appStore.fetchState();
 
 // Connection status

dashboard/src/routes/+page.svelte

@@ -42,9 +42,6 @@
     setSelectedChatModel,
     selectedChatModel,
     sendMessage,
-    generateImage,
-    editImage,
-    editingImage,
     messages,
     debugMode,
     toggleDebugMode,
@@ -52,12 +49,6 @@
     toggleTopologyOnlyMode,
     chatSidebarVisible,
     toggleChatSidebarVisible,
-    mobileChatSidebarOpen,
-    toggleMobileChatSidebar,
-    setMobileChatSidebarOpen,
-    mobileRightSidebarOpen,
-    toggleMobileRightSidebar,
-    setMobileRightSidebarOpen,
     nodeThunderbolt,
     nodeRdmaCtl,
     thunderboltBridgeCycles,
@@ -88,8 +79,6 @@
   const debugEnabled = $derived(debugMode());
   const topologyOnlyEnabled = $derived(topologyOnlyMode());
   const sidebarVisible = $derived(chatSidebarVisible());
-  const mobileChatOpen = $derived(mobileChatSidebarOpen());
-  const mobileRightOpen = $derived(mobileRightSidebarOpen());
   const tbBridgeCycles = $derived(thunderboltBridgeCycles());
   const tbBridgeData = $derived(nodeThunderboltBridge());
   const identitiesData = $derived(nodeIdentities());
@@ -837,52 +826,6 @@
     if (!model?.tasks) return false;
     return model.tasks.includes("ImageToImage");
   }
-
-  // Route a message to the correct endpoint based on model capabilities.
-  // Image models go to generateImage/editImage; text models go to sendMessage.
-  function routeMessage(
-    content: string,
-    files?: {
-      id: string;
-      name: string;
-      type: string;
-      textContent?: string;
-      preview?: string;
-    }[],
-  ) {
-    const model = selectedChatModel();
-    if (!model) {
-      sendMessage(content, files, null);
-      return;
-    }
-
-    const currentEditImage = editingImage();
-
-    // Image editing mode (explicit edit or attached image with ImageToImage model)
-    if (currentEditImage && content && modelSupportsImageEditing(model)) {
-      editImage(content, currentEditImage.imageDataUrl);
-      return;
-    }
-    if (
-      modelSupportsImageEditing(model) &&
-      files?.length &&
-      files[0].preview &&
-      content
-    ) {
-      editImage(content, files[0].preview);
-      return;
-    }
-
-    // Text-to-image generation
-    if (modelSupportsImageGeneration(model) && content) {
-      generateImage(content);
-      return;
-    }
-
-    // Default: text chat
-    sendMessage(content, files, null);
-  }
-
   let selectedSharding = $state<"Pipeline" | "Tensor">("Pipeline");
   type InstanceMeta = "MlxRing" | "MlxJaccl";
 
@@ -1576,11 +1519,7 @@
           downloadKind
         ] as Record<string, unknown>;
 
-        if (
-          downloadKind !== "DownloadOngoing" &&
-          downloadKind !== "DownloadPending"
-        )
-          continue;
+        if (downloadKind !== "DownloadOngoing") continue;
         if (!downloadPayload) continue;
 
         const downloadModelId = extractModelIdFromDownload(downloadPayload);
@@ -1595,38 +1534,9 @@
           if (downloadModelId !== modelId) continue;
         }
 
-        // For DownloadPending with partial bytes (paused/resumed downloads),
-        // synthesize a progress object from the top-level downloaded/total fields
-        let progress: DownloadProgress | null;
-        if (downloadKind === "DownloadPending") {
-          const pendingDownloaded = getBytes(
-            downloadPayload.downloaded ??
-              downloadPayload.downloaded_bytes ??
-              downloadPayload.downloadedBytes,
-          );
-          const pendingTotal = getBytes(
-            downloadPayload.total ??
-              downloadPayload.total_bytes ??
-              downloadPayload.totalBytes,
-          );
-          if (pendingDownloaded <= 0 && pendingTotal <= 0) continue;
-          isDownloading = true;
-          progress = {
-            totalBytes: pendingTotal,
-            downloadedBytes: pendingDownloaded,
-            speed: 0,
-            etaMs: 0,
-            percentage:
-              pendingTotal > 0 ? (pendingDownloaded / pendingTotal) * 100 : 0,
-            completedFiles: 0,
-            totalFiles: 0,
-            files: [],
-          };
-        } else {
-          isDownloading = true;
-          progress = parseDownloadProgress(downloadPayload);
-        }
+        isDownloading = true;
 
+        const progress = parseDownloadProgress(downloadPayload);
         if (progress) {
           // Sum all values across nodes - each node downloads independently
           totalBytes += progress.totalBytes;
@@ -1778,11 +1688,7 @@
           }
         }
 
-        if (
-          downloadKind !== "DownloadOngoing" &&
-          downloadKind !== "DownloadPending"
-        )
-          continue;
+        if (downloadKind !== "DownloadOngoing") continue;
         if (!downloadPayload) continue;
 
         // Check if this download is for this instance's model
@@ -1792,37 +1698,9 @@
           downloadModelId &&
           downloadModelId === instanceModelId
         ) {
-          // For DownloadPending with partial bytes, synthesize progress
-          let progress: DownloadProgress | null;
-          if (downloadKind === "DownloadPending") {
-            const pendingDownloaded = getBytes(
-              downloadPayload.downloaded ??
-                downloadPayload.downloaded_bytes ??
-                downloadPayload.downloadedBytes,
-            );
-            const pendingTotal = getBytes(
-              downloadPayload.total ??
-                downloadPayload.total_bytes ??
-                downloadPayload.totalBytes,
-            );
-            if (pendingDownloaded <= 0 && pendingTotal <= 0) continue;
-            isDownloading = true;
-            progress = {
-              totalBytes: pendingTotal,
-              downloadedBytes: pendingDownloaded,
-              speed: 0,
-              etaMs: 0,
-              percentage:
-                pendingTotal > 0 ? (pendingDownloaded / pendingTotal) * 100 : 0,
-              completedFiles: 0,
-              totalFiles: 0,
-              files: [],
-            };
-          } else {
-            isDownloading = true;
-            progress = parseDownloadProgress(downloadPayload);
-          }
+          isDownloading = true;
 
+          const progress = parseDownloadProgress(downloadPayload);
           if (progress) {
             // Sum all values across nodes - each node downloads independently
             totalBytes += progress.totalBytes;
@@ -2900,7 +2778,7 @@
         // Running model is same or better tier — use it directly
         setSelectedChatModel(bestRunning.id);
         if (!chatStarted) createConversation();
-        routeMessage(content, files);
+        sendMessage(content, files);
         return;
       }
     }
@@ -2917,7 +2795,7 @@
     if (hasRunningInstance(autoModel.id)) {
       setSelectedChatModel(autoModel.id);
       if (!chatStarted) createConversation();
-      routeMessage(content, files);
+      sendMessage(content, files);
       return;
     }
 
@@ -3070,7 +2948,7 @@
       if (pendingAutoMessage) {
         const msg = pendingAutoMessage;
         pendingAutoMessage = null;
-        routeMessage(msg.content, msg.files);
+        sendMessage(msg.content, msg.files);
       }
       return;
     }
@@ -3149,7 +3027,7 @@
     // Model is selected and running — send directly
     if (model && hasRunningInstance(model)) {
       chatLaunchState = "ready";
-      routeMessage(content, files);
+      sendMessage(content, files, null);
       return;
     }
 
@@ -4726,35 +4604,16 @@
       showSidebarToggle={true}
       {sidebarVisible}
       onToggleSidebar={toggleChatSidebarVisible}
-      showMobileMenuToggle={true}
-      mobileMenuOpen={mobileChatOpen}
-      onToggleMobileMenu={toggleMobileChatSidebar}
-      showMobileRightToggle={!chatStarted && !topologyOnlyEnabled}
-      {mobileRightOpen}
-      onToggleMobileRight={toggleMobileRightSidebar}
       downloadProgress={activeDownloadSummary}
     />
   {/if}
 
-  <!-- Mobile Chat Sidebar Drawer -->
-  {#if !topologyOnlyEnabled}
-    <ChatSidebar
-      isMobileDrawer={true}
-      isOpen={mobileChatOpen}
-      onClose={() => setMobileChatSidebarOpen(false)}
-      onNewChat={handleNewChat}
-      onSelectConversation={() => {
-        userForcedIdle = false;
-      }}
-    />
-  {/if}
-
   <!-- Main Content -->
   <main class="flex-1 flex overflow-hidden relative">
-    <!-- Left: Conversation History Sidebar (hidden in topology-only mode, welcome state, or when toggled off) - Desktop only -->
+    <!-- Left: Conversation History Sidebar (hidden in topology-only mode, welcome state, or when toggled off) -->
     {#if !topologyOnlyEnabled && sidebarVisible}
       <div
-        class="hidden md:block w-80 flex-shrink-0 border-r border-exo-yellow/10"
+        class="w-80 flex-shrink-0 border-r border-exo-yellow/10"
         role="complementary"
         aria-label="Conversation history"
       >
@@ -5064,33 +4923,11 @@
           </div>
         </div>
 
-        <!-- Mobile Right Sidebar Drawer (Instances) -->
-        {#if mobileRightOpen}
-          <!-- Overlay backdrop -->
-          <button
-            type="button"
-            class="fixed inset-0 bg-black/60 backdrop-blur-sm z-40 md:hidden"
-            onclick={() => setMobileRightSidebarOpen(false)}
-            aria-label="Close instances panel"
-          ></button>
-          <!-- Drawer panel -->
-          <aside
-            class="fixed right-0 top-0 bottom-0 w-80 bg-exo-dark-gray border-l border-exo-yellow/10 z-50 flex flex-col md:hidden overflow-y-auto"
-            aria-label="Instance controls mobile"
-          >
-            {@render rightSidebarContent()}
-          </aside>
-        {/if}
-
-        <!-- Right Sidebar: Instance Controls (wider on welcome page for better visibility) - Desktop only -->
+        <!-- Right Sidebar: Instance Controls (wider on welcome page for better visibility) -->
         <aside
-          class="hidden md:flex w-80 border-l border-exo-yellow/10 bg-exo-dark-gray flex-col flex-shrink-0"
+          class="w-80 border-l border-exo-yellow/10 bg-exo-dark-gray flex flex-col flex-shrink-0"
           aria-label="Instance controls"
         >
-          {@render rightSidebarContent()}
-        </aside>
-
-        {#snippet rightSidebarContent()}
           <!-- Running Instances Panel (only shown when instances exist) - Scrollable -->
           {#if instanceCount > 0}
             <div class="p-4 flex-shrink-0">
@@ -5972,7 +5809,7 @@
               {/if}
             </div>
           </div>
-        {/snippet}
+        </aside>
       </div>
     {:else}
       <!-- CHAT STATE: Chat + Mini-Map -->
@@ -6185,10 +6022,10 @@
           {/if}
         </div>
 
-        <!-- Right: Mini-Map Sidebar - Desktop only -->
+        <!-- Right: Mini-Map Sidebar -->
         {#if minimized}
           <aside
-            class="hidden md:flex w-80 border-l border-exo-yellow/20 bg-exo-dark-gray flex-col flex-shrink-0 overflow-y-auto"
+            class="w-80 border-l border-exo-yellow/20 bg-exo-dark-gray flex flex-col flex-shrink-0 overflow-y-auto"
             in:fly={{ x: 100, duration: 400, easing: cubicInOut }}
             aria-label="Cluster topology"
           >

flake.nix

@@ -112,20 +112,17 @@
             };
           };
 
-          packages = {
-            babeld = pkgs.callPackage ./nix/babeld.nix { };
-          } // lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin (
+          packages = lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin (
             let
               uvLock = builtins.fromTOML (builtins.readFile ./uv.lock);
               mlxPackage = builtins.head (builtins.filter (p: p.name == "mlx" && p.source ? git) uvLock.package);
               uvLockMlxVersion = mlxPackage.version;
-              uvLockMlxRev = builtins.elemAt (builtins.split "#" mlxPackage.source.git) 2;
             in
             {
               metal-toolchain = pkgs.callPackage ./nix/metal-toolchain.nix { };
               mlx = pkgs.callPackage ./nix/mlx.nix {
                 inherit (self'.packages) metal-toolchain;
-                inherit uvLockMlxVersion uvLockMlxRev;
+                inherit uvLockMlxVersion;
               };
               default = self'.packages.exo;
             }
@@ -159,6 +156,9 @@
                 just
                 jq
               ]
+              ++ lib.optionals stdenv.isLinux [
+                unixtools.ifconfig
+              ]
               ++ lib.optionals stdenv.isDarwin [
                 macmon
               ];

nix/babeld.nix

@@ -1,25 +0,0 @@
-{ stdenv
-, lib
-, fetchurl
-}:
-stdenv.mkDerivation rec {
-  pname = "babeld";
-  version = "1.13.1";
-
-  src = fetchurl {
-    url = "https://www.irif.fr/~jch/software/files/${pname}-${version}.tar.gz";
-    hash = "sha256-FfJNJtoMz8Bzq83vAwnygeRoTyqnESb4JlcsTIRejdk=";
-  };
-
-  outputs = [
-    "out"
-    "man"
-  ];
-
-  makeFlags = [
-    "PREFIX=${placeholder "out"}"
-    "ETCDIR=${placeholder "out"}/etc"
-  ]
-  ++ lib.optional stdenv.isDarwin "LDLIBS=''";
-}
-

nix/mlx.nix

@@ -11,7 +11,6 @@
 , fmt
 , python313Packages
 , uvLockMlxVersion
-, uvLockMlxRev
 }:
 
 assert stdenv.isDarwin;
@@ -42,13 +41,15 @@ let
 
   mlx = stdenv.mkDerivation rec {
     pname = "mlx";
-    version = uvLockMlxVersion;
+    version = let v = "0.30.7.dev20260225+257d5692"; in
+      assert v == uvLockMlxVersion || throw "MLX version mismatch: nix/mlx.nix has ${v} but uv.lock has ${uvLockMlxVersion}. Update both the version and hash in nix/mlx.nix.";
+      v;
     pyproject = true;
 
     src = fetchFromGitHub {
       owner = "rltakashige";
       repo = "mlx-jaccl-fix-small-recv";
-      rev = uvLockMlxRev;
+      rev = "257d5692fc7af6bba3b8afaeb63c549b7d1e43d5";
       hash = "sha256-GosFIWxIB48Egb1MqJrR3xhsUsQeWdRk5rV93USY6wQ=";
     };
 

pyproject.toml

@@ -62,7 +62,7 @@ members = ["rust/exo_pyo3_bindings", "bench"]
 [tool.uv.sources]
 exo_pyo3_bindings = { workspace = true }
 mlx = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git", branch = "address-rdma-gpu-locks", marker = "sys_platform == 'darwin'" }
-mlx-lm = { git = "https://github.com/rltakashige/mlx-lm", branch = "leo/eval-left-padding-in-batched-rotation" }
+mlx-lm = { git = "https://github.com/ml-explore/mlx-lm", rev = "834fac934c4e04de9b3d723e2b9287a2c60cfd4a" }
 # Uncomment to use local mlx/mlx-lm development versions:
 # mlx = { path = "/Users/Shared/mlx", editable=true }
 # mlx-lm = { path = "/Users/Shared/mlx-lm", editable=true }
@@ -125,7 +125,6 @@ extend-exclude = [
   "shared/protobufs/**",
   "*mlx_typings/**",
   "rust/exo_pyo3_bindings/**",
-  "bench/vendor/**",
 ]
 
 [tool.ruff.lint]

python/parts.nix

@@ -43,27 +43,6 @@
             final.setuptools
           ];
         });
-        # rouge-score and sacrebleu don't declare setuptools as a build dependency
-        rouge-score = prev.rouge-score.overrideAttrs (old: {
-          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
-            final.setuptools
-          ];
-        });
-        sacrebleu = prev.sacrebleu.overrideAttrs (old: {
-          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
-            final.setuptools
-          ];
-        });
-        sqlitedict = prev.sqlitedict.overrideAttrs (old: {
-          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
-            final.setuptools
-          ];
-        });
-        word2number = prev.word2number.overrideAttrs (old: {
-          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
-            final.setuptools
-          ];
-        });
       } // lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin {
         # Use our pure Nix-built MLX with Metal support (macOS only)
         mlx = self'.packages.mlx;
@@ -117,16 +96,13 @@
         "lib/python3.13/site-packages/nvidia*"
       ];
 
-      # Exclude bench deps from main env (bench has its own benchVenv)
-      exoDeps = removeAttrs workspace.deps.default [ "exo-bench" ];
-
-      exoVenv = (pythonSet.mkVirtualEnv "exo-env" exoDeps).overrideAttrs {
+      exoVenv = (pythonSet.mkVirtualEnv "exo-env" workspace.deps.default).overrideAttrs {
         venvIgnoreCollisions = venvCollisionPaths;
       };
 
       # Virtual environment with dev dependencies for testing
       testVenv = (pythonSet.mkVirtualEnv "exo-test-env" (
-        exoDeps // {
+        workspace.deps.default // {
           exo = [ "dev" ]; # Include pytest, pytest-asyncio, pytest-env
         }
       )).overrideAttrs {
@@ -182,10 +158,8 @@
           exo-test-env = testVenv;
         } // {
         exo-bench = mkBenchScript "exo-bench" (inputs.self + /bench/exo_bench.py);
-        exo-eval = mkBenchScript "exo-eval" (inputs.self + /bench/exo_eval.py);
         exo-eval-tool-calls = mkBenchScript "exo-eval-tool-calls" (inputs.self + /bench/eval_tool_calls.py);
         exo-get-all-models-on-cluster = mkSimplePythonScript "exo-get-all-models-on-cluster" (inputs.self + /tests/get_all_models_on_cluster.py);
-        exo-mlx-bandwidth-test = mkPythonScript "exo-mlx-bandwidth-test" (inputs.self + /bench/test_mlx_bandwidth.py);
       };
 
       checks = {

resources/inference_model_cards/mlx-community--DeepSeek-V3.1-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/DeepSeek-V3.1-4bit"
 n_layers = 61
 hidden_size = 7168
-num_key_value_heads = 128
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "deepseek"

resources/inference_model_cards/mlx-community--DeepSeek-V3.1-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/DeepSeek-V3.1-8bit"
 n_layers = 61
 hidden_size = 7168
-num_key_value_heads = 128
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "deepseek"

resources/inference_model_cards/mlx-community--GLM-4.5-Air-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.5-Air-8bit"
 n_layers = 46
 hidden_size = 4096
-num_key_value_heads = 8
 supports_tensor = false
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.5-Air-bf16.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.5-Air-bf16"
 n_layers = 46
 hidden_size = 4096
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.7-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.7-4bit"
 n_layers = 91
 hidden_size = 5120
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.7-6bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.7-6bit"
 n_layers = 91
 hidden_size = 5120
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.7-8bit-gs32.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.7-8bit-gs32"
 n_layers = 91
 hidden_size = 5120
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.7-Flash-4bit"
 n_layers = 47
 hidden_size = 2048
-num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-5bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.7-Flash-5bit"
 n_layers = 47
 hidden_size = 2048
-num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-6bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.7-Flash-6bit"
 n_layers = 47
 hidden_size = 2048
-num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-4.7-Flash-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-4.7-Flash-8bit"
 n_layers = 47
 hidden_size = 2048
-num_key_value_heads = 20
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-5-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-5-8bit-MXFP8"
 n_layers = 78
 hidden_size = 6144
-num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-5-MXFP4-Q8.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-5-MXFP4-Q8"
 n_layers = 78
 hidden_size = 6144
-num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--GLM-5-bf16.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/GLM-5"
 n_layers = 78
 hidden_size = 6144
-num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "glm"

resources/inference_model_cards/mlx-community--Kimi-K2-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Kimi-K2-Instruct-4bit"
 n_layers = 61
 hidden_size = 7168
-num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "kimi"

resources/inference_model_cards/mlx-community--Kimi-K2-Thinking.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Kimi-K2-Thinking"
 n_layers = 61
 hidden_size = 7168
-num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "kimi"

resources/inference_model_cards/mlx-community--Kimi-K2.5.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Kimi-K2.5"
 n_layers = 61
 hidden_size = 7168
-num_key_value_heads = 64
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "kimi"

resources/inference_model_cards/mlx-community--Llama-3.1-Nemotron-70B-Instruct-HF-4bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/Llama-3.1-Nemotron-70B-Instruct-HF-4bit"
-n_layers = 80
-hidden_size = 8192
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "llama"
-quantization = "4bit"
-base_model = "NVIDIA Llama-3.1-Nemotron-70B-Instruct"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 39688355840

resources/inference_model_cards/mlx-community--Llama-3.1-Nemotron-70B-Instruct-HF-8bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/Llama-3.1-Nemotron-70B-Instruct-HF-8bit"
-n_layers = 80
-hidden_size = 8192
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "llama"
-quantization = "8bit"
-base_model = "NVIDIA Llama-3.1-Nemotron-70B-Instruct"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 74964549632

resources/inference_model_cards/mlx-community--Llama-3.1-Nemotron-70B-Instruct-HF-bf16.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/Llama-3.1-Nemotron-70B-Instruct-HF-bf16"
-n_layers = 80
-hidden_size = 8192
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "llama"
-quantization = "bf16"
-base_model = "NVIDIA Llama-3.1-Nemotron-70B-Instruct"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 141107412992

resources/inference_model_cards/mlx-community--Llama-3.1-Nemotron-Nano-4B-v1.1-4bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/Llama-3.1-Nemotron-Nano-4B-v1.1-4bit"
-n_layers = 32
-hidden_size = 3072
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "llama"
-quantization = "4bit"
-base_model = "NVIDIA Llama-3.1-Nemotron-Nano-4B-v1.1"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 2538706944

resources/inference_model_cards/mlx-community--Llama-3.1-Nemotron-Nano-4B-v1.1-8bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/Llama-3.1-Nemotron-Nano-4B-v1.1-8bit"
-n_layers = 32
-hidden_size = 3072
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "llama"
-quantization = "8bit"
-base_model = "NVIDIA Llama-3.1-Nemotron-Nano-4B-v1.1"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 4794980352

resources/inference_model_cards/mlx-community--Llama-3.1-Nemotron-Nano-4B-v1.1-bf16.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/Llama-3.1-Nemotron-Nano-4B-v1.1-bf16"
-n_layers = 32
-hidden_size = 3072
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "llama"
-quantization = "bf16"
-base_model = "NVIDIA Llama-3.1-Nemotron-Nano-4B-v1.1"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 9025492992

resources/inference_model_cards/mlx-community--Llama-3.2-1B-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Llama-3.2-1B-Instruct-4bit"
 n_layers = 16
 hidden_size = 2048
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Llama-3.2-3B-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Llama-3.2-3B-Instruct-4bit"
 n_layers = 28
 hidden_size = 3072
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Llama-3.2-3B-Instruct-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Llama-3.2-3B-Instruct-8bit"
 n_layers = 28
 hidden_size = 3072
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Llama-3.3-70B-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Llama-3.3-70B-Instruct-4bit"
 n_layers = 80
 hidden_size = 8192
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Llama-3.3-70B-Instruct-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Llama-3.3-70B-Instruct-8bit"
 n_layers = 80
 hidden_size = 8192
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-70B-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Meta-Llama-3.1-70B-Instruct-4bit"
 n_layers = 80
 hidden_size = 8192
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-8B-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Meta-Llama-3.1-8B-Instruct-4bit"
 n_layers = 32
 hidden_size = 4096
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-8B-Instruct-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Meta-Llama-3.1-8B-Instruct-8bit"
 n_layers = 32
 hidden_size = 4096
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--Meta-Llama-3.1-8B-Instruct-bf16.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"
 n_layers = 32
 hidden_size = 4096
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "llama"

resources/inference_model_cards/mlx-community--MiniMax-M2.1-3bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/MiniMax-M2.1-3bit"
 n_layers = 61
 hidden_size = 3072
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "minimax"

resources/inference_model_cards/mlx-community--MiniMax-M2.1-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/MiniMax-M2.1-8bit"
 n_layers = 61
 hidden_size = 3072
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "minimax"

resources/inference_model_cards/mlx-community--MiniMax-M2.5-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/MiniMax-M2.5-4bit"
 n_layers = 62
 hidden_size = 3072
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "minimax"

resources/inference_model_cards/mlx-community--MiniMax-M2.5-6bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/MiniMax-M2.5-6bit"
 n_layers = 62
 hidden_size = 3072
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "minimax"

resources/inference_model_cards/mlx-community--MiniMax-M2.5-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/MiniMax-M2.5-8bit"
 n_layers = 62
 hidden_size = 3072
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "minimax"

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-4Bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-4Bit"
-n_layers = 52
-hidden_size = 2688
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "4bit"
-base_model = "NVIDIA Nemotron-3-Nano-30B-A3B"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 17775342336

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-5Bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-5Bit"
-n_layers = 52
-hidden_size = 2688
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "5bit"
-base_model = "NVIDIA Nemotron-3-Nano-30B-A3B"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 21721476864

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-6Bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-6Bit"
-n_layers = 52
-hidden_size = 2688
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "6bit"
-base_model = "NVIDIA Nemotron-3-Nano-30B-A3B"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 25667611392

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-8Bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-8Bit"
-n_layers = 52
-hidden_size = 2688
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "8bit"
-base_model = "NVIDIA Nemotron-3-Nano-30B-A3B"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 33559880448

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-BF16.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-BF16"
-n_layers = 52
-hidden_size = 2688
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "bf16"
-base_model = "NVIDIA Nemotron-3-Nano-30B-A3B"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 63155889408

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-MXFP4.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-MXFP4"
-n_layers = 52
-hidden_size = 2688
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "4bit"
-base_model = "NVIDIA Nemotron-3-Nano-30B-A3B"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 16788808704

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4"
-n_layers = 52
-hidden_size = 2688
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "4bit"
-base_model = "NVIDIA Nemotron-3-Nano-30B-A3B"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 19323906944

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-Nano-9B-v2-4bits.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-Nano-9B-v2-4bits"
-n_layers = 56
-hidden_size = 4480
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "4bit"
-base_model = "NVIDIA Nemotron-Nano-9B-v2"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 5002791936

resources/inference_model_cards/mlx-community--NVIDIA-Nemotron-Nano-9B-v2-6bit.toml

@@ -1,12 +0,0 @@
-model_id = "mlx-community/NVIDIA-Nemotron-Nano-9B-v2-6bit"
-n_layers = 56
-hidden_size = 4480
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "nemotron"
-quantization = "6bit"
-base_model = "NVIDIA Nemotron-Nano-9B-v2"
-capabilities = ["text"]
-
-[storage_size]
-in_bytes = 7224298496

resources/inference_model_cards/mlx-community--Qwen3-0.6B-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-0.6B-4bit"
 n_layers = 28
 hidden_size = 1024
-num_key_value_heads = 8
 supports_tensor = false
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-0.6B-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-0.6B-8bit"
 n_layers = 28
 hidden_size = 1024
-num_key_value_heads = 8
 supports_tensor = false
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-235B-A22B-Instruct-2507-4bit"
 n_layers = 94
 hidden_size = 4096
-num_key_value_heads = 4
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-235B-A22B-Instruct-2507-8bit"
 n_layers = 94
 hidden_size = 4096
-num_key_value_heads = 4
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-30B-A3B-4bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 4
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-30B-A3B-8bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 4
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Coder-480B-A35B-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit"
 n_layers = 62
 hidden_size = 6144
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Coder-480B-A35B-Instruct-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Coder-480B-A35B-Instruct-8bit"
 n_layers = 62
 hidden_size = 6144
-num_key_value_heads = 8
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Coder-Next-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Coder-Next-4bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Coder-Next-5bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Coder-Next-5bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Coder-Next-6bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Coder-Next-6bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Coder-Next-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Coder-Next-8bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Coder-Next-bf16.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Coder-Next-bf16"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Instruct-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Instruct-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3-Next-80B-A3B-Thinking-8bit"
 n_layers = 48
 hidden_size = 2048
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-122B-A10B-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3.5-122B-A10B-4bit"
 n_layers = 48
 hidden_size = 3072
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-122B-A10B-6bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3.5-122B-A10B-6bit"
 n_layers = 48
 hidden_size = 3072
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-122B-A10B-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3.5-122B-A10B-8bit"
 n_layers = 48
 hidden_size = 3072
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-122B-A10B-bf16.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3.5-122B-A10B-bf16"
 n_layers = 48
 hidden_size = 3072
-num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-27B-4bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3.5-27B-4bit"
 n_layers = 64
 hidden_size = 5120
-num_key_value_heads = 4
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-27B-8bit.toml

@@ -1,7 +1,6 @@
 model_id = "mlx-community/Qwen3.5-27B-8bit"
 n_layers = 64
 hidden_size = 5120
-num_key_value_heads = 4
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"