Call warmup_speculative at startup to pre-compile LpB kernels

The warmup_speculative() function was defined but never called.
Custom Metal kernels (LpB) require first-call compilation (~200ms).
Without warmup, the first speculative cycle is slow, dragging down
average TPS by 10-20% on short generations.

In mlx_bench testing: cold 48 TPS → warm 60 TPS for DFlash,
cold 39 TPS → warm 44 TPS for MTP.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

.mlx_typings/mlx/core/__init__.pyi

@@ -2396,7 +2396,7 @@ def degrees(a: array, /, *, stream: Stream | Device | None = ...) -> array:
         array: The angles in degrees.
     """
 
-def depends(inputs: array | Sequence[array], dependencies: array | Sequence[array]):
+def depends[T](inputs: T, dependencies: array | Sequence[array]) -> T:
     """
     Insert dependencies between arrays in the graph. The outputs are
     identical to ``inputs`` but with dependencies on ``dependencies``.

.mlx_typings/mlx/nn/__init__.pyi

@@ -1,9 +1,5 @@
-"""
-This type stub file was generated by pyright.
-"""
-
-from layers import *
-from utils import *
+from .layers import *
+from .utils import *
 
 from . import init as init
 from . import losses as losses

.mlx_typings/mlx/nn/layers/__init__.pyi

@@ -1,20 +1,16 @@
-"""
-This type stub file was generated by pyright.
-"""
-
-from activations import *
-from base import *
-from containers import *
-from convolution import *
-from convolution_transpose import *
-from distributed import *
-from dropout import *
-from embedding import *
-from linear import *
-from normalization import *
-from pooling import *
-from positional_encoding import *
-from quantized import *
-from recurrent import *
-from transformer import *
-from upsample import *
+from .activations import *
+from .base import *
+from .containers import *
+from .convolution import *
+from .convolution_transpose import *
+from .distributed import *
+from .dropout import *
+from .embedding import *
+from .linear import *
+from .normalization import *
+from .pooling import *
+from .positional_encoding import *
+from .quantized import *
+from .recurrent import *
+from .transformer import *
+from .upsample import *

.mlx_typings/mlx/nn/layers/base.pyi

@@ -53,7 +53,7 @@ class Module(dict):
         mx.eval(model.parameters())
     """
 
-    __call__: Callable
+    def __call__(self, *args: Any, **kwargs: Any) -> mx.array: ...
     def __init__(self) -> None:
         """Should be called by the subclasses of ``Module``."""
 

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

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

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

@@ -40,6 +40,10 @@ 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

@@ -88,6 +88,9 @@ 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

@@ -30,7 +30,7 @@ def str2bool(string):  # -> bool:
 def setup_arg_parser():  # -> ArgumentParser:
     """Set up and return the argument parser."""
 
-generation_stream = ...
+generation_stream: mx.Stream
 
 @contextlib.contextmanager
 def wired_limit(
@@ -266,12 +266,12 @@ def _merge_caches(caches: Any) -> List[Any]: ...
 class Batch:
     uids: List[int]
     y: mx.array
-    logprobs: mx.array
+    logprobs: List[mx.array] | mx.array
     max_tokens: List[int]
     num_tokens: List[int]
     cache: List[Any]
-    samplers: List[Any]
-    logits_processors: List[Any]
+    samplers: List[Callable[[mx.array], mx.array] | None]
+    logits_processors: List[List[Callable[[mx.array, mx.array], mx.array]]]
     tokens: List[mx.array]
     def __len__(self) -> int: ...
     def filter(self, keep_idx: List[int]) -> None: ...
@@ -279,13 +279,18 @@ class Batch:
     def extract_cache(self, idx: int) -> List[Any]: ...
 
 class BatchGenerator:
-    model: Any
+    model: nn.Module
+    sampler: Callable[[mx.array], mx.array]
+    stop_tokens: set[int]
     max_kv_size: Optional[int]
     prefill_step_size: int
+    completion_batch_size: int
+    prefill_batch_size: int
     unprocessed_prompts: List[Any]
     active_batch: Optional[Batch]
     prompt_progress_callback: Callable[[List[Tuple[int, int, int]]], None]
     _stats: BatchStats
+    _next_count: int
 
     @dataclass
     class Response:

.mlx_typings/mlx_lm/models/cache.pyi

@@ -88,8 +88,8 @@ def create_attention_mask(
 ) -> array | Literal["causal"] | None: ...
 
 class _BaseCache(Cache):
-    keys: mx.array
-    values: mx.array
+    keys: mx.array | None
+    values: mx.array | None
     offset: int
     @property
     def state(self) -> tuple[mx.array | None, mx.array | None]: ...
@@ -268,29 +268,14 @@ class CacheList(_BaseCache):
         """
 
 class BatchKVCache(_BaseCache):
-    step = ...
-    def __init__(self, left_padding: List[int]) -> None:
-        """
-        The BatchKV cache expects inputs to be left-padded.
-
-        E.g. the following prompts:
-
-            [1, 3, 5]
-            [7]
-            [2, 6, 8, 9]
-
-        Should be padded like so:
-
-            [0, 1, 3, 5]
-            [0, 0, 0, 7]
-            [2, 6, 8, 9]
-
-        And ``left_padding`` specifies the amount of padding for each.
-        In this case, ``left_padding = [1, 3, 0]``.
-        """
-
-    def update_and_fetch(self, keys, values):  # -> tuple[array | Any, array | Any]:
-        ...
+    step: int
+    keys: array | None
+    values: array | None
+    offset: array
+    left_padding: array
+    _idx: int
+    def __init__(self, left_padding: List[int]) -> None: ...
+    def update_and_fetch(self, keys: array, values: array) -> tuple[array, array]: ...
     @property
     def state(
         self,
@@ -316,12 +301,21 @@ class BatchKVCache(_BaseCache):
         """
 
 class BatchRotatingKVCache(_BaseCache):
-    step = ...
-    def __init__(self, max_size, left_padding: List[int]) -> None: ...
-    def update_and_fetch(
-        self, keys, values
-    ):  # -> tuple[array | Any, array | Any] | tuple[array | Any, array | Any | None]:
-        ...
+    step: int
+    keys: array | None
+    values: array | None
+    offset: array
+    left_padding: array
+    max_size: int
+    _idx: int
+    _offset: int
+    rotated: bool
+    _lengths: array | None
+    def __init__(self, max_size: int, left_padding: List[int]) -> None: ...
+    def _trim(self, trim_size: int, v: array, append: array | None = ...) -> array: ...
+    def _update_in_place(self, keys: array, values: array) -> tuple[array, array]: ...
+    def _update_concat(self, keys: array, values: array) -> tuple[array, array]: ...
+    def update_and_fetch(self, keys: array, values: array) -> tuple[array, array]: ...
     @property
     def state(
         self,

.mlx_typings/mlx_lm/models/gated_delta.pyi

@@ -0,0 +1,35 @@
+from typing import Optional
+
+import mlx.core as mx
+
+def compute_g(A_log: mx.array, a: mx.array, dt_bias: mx.array) -> mx.array: ...
+def gated_delta_update(
+    q: mx.array,
+    k: mx.array,
+    v: mx.array,
+    a: mx.array,
+    b: mx.array,
+    A_log: mx.array,
+    dt_bias: mx.array,
+    state: Optional[mx.array] = ...,
+    mask: Optional[mx.array] = ...,
+    use_kernel: bool = ...,
+) -> tuple[mx.array, mx.array]: ...
+def gated_delta_ops(
+    q: mx.array,
+    k: mx.array,
+    v: mx.array,
+    g: mx.array,
+    beta: mx.array,
+    state: Optional[mx.array] = ...,
+    mask: Optional[mx.array] = ...,
+) -> tuple[mx.array, mx.array]: ...
+def gated_delta_kernel(
+    q: mx.array,
+    k: mx.array,
+    v: mx.array,
+    g: mx.array,
+    beta: mx.array,
+    state: mx.array,
+    mask: Optional[mx.array] = ...,
+) -> tuple[mx.array, mx.array]: ...

.mlx_typings/mlx_lm/models/nemotron_h.pyi

@@ -0,0 +1,154 @@
+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

@@ -5,6 +5,7 @@ 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):
@@ -99,6 +100,8 @@ 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__(
@@ -121,3 +124,4 @@ 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/rope_utils.pyi

@@ -0,0 +1,51 @@
+from typing import Any, Optional
+
+import mlx.nn as nn
+
+class YarnRoPE(nn.Module):
+    def __init__(
+        self,
+        dims: int,
+        traditional: bool = ...,
+        max_position_embeddings: int = ...,
+        base: float = ...,
+        scaling_factor: float = ...,
+        original_max_position_embeddings: int = ...,
+        beta_fast: float = ...,
+        beta_slow: float = ...,
+        mscale: float = ...,
+        mscale_all_dim: float = ...,
+    ) -> None: ...
+
+class Llama3RoPE(nn.Module):
+    def __init__(
+        self,
+        dims: int,
+        traditional: bool = ...,
+        max_position_embeddings: int = ...,
+        base: float = ...,
+        scaling_factor: float = ...,
+        original_max_position_embeddings: int = ...,
+        low_freq_factor: float = ...,
+        high_freq_factor: float = ...,
+    ) -> None: ...
+
+class SuScaledRoPE(nn.Module):
+    def __init__(
+        self,
+        dims: int,
+        traditional: bool = ...,
+        max_position_embeddings: int = ...,
+        base: float = ...,
+        short_factor: Any = ...,
+        long_factor: Any = ...,
+        original_max_position_embeddings: int = ...,
+    ) -> None: ...
+
+def initialize_rope(
+    dims: int,
+    base: float = ...,
+    traditional: bool = ...,
+    scaling_config: Optional[dict[str, Any]] = ...,
+    max_position_embeddings: Optional[int] = ...,
+) -> nn.Module: ...

.mlx_typings/mlx_lm/models/switch_layers.pyi

@@ -73,6 +73,9 @@ 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): ...
-    def add_token(self, token): ...
-    def finalize(self): ...
+    def reset(self) -> None: ...
+    def add_token(self, token: int) -> None: ...
+    def finalize(self) -> None: ...
     @property
-    def last_segment(self):
+    def last_segment(self) -> str:
         """Return the last segment of readable text since last time this property was accessed."""
 
 class NaiveStreamingDetokenizer(StreamingDetokenizer):

bench/exo_bench.py

@@ -496,20 +496,26 @@ def main() -> int:
                                 all_rows.append(row)
 
                             if batch_results:
-                                agg_gen_tps = sum(
+                                valid_gen_tps = [
                                     x["stats"]["generation_tps"]
                                     for x, _ in batch_results
                                     if x["stats"]["generation_tps"] > 0
+                                ]
+                                per_req_tps = (
+                                    mean(valid_gen_tps) if valid_gen_tps else 0.0
                                 )
+                                agg_gen_tps = per_req_tps * concurrency
                                 logger.info(
                                     f"[concurrent {concurrency}x]  "
                                     f"agg_gen_tps={agg_gen_tps:.2f}  "
+                                    f"per_req_tps={per_req_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"] for x in runs)
+                        per_req_tps = mean(x["stats"]["generation_tps"] for x in runs)
+                        gen_tps = per_req_tps * concurrency
                         ptok = mean(x["stats"]["prompt_tokens"] for x in runs)
                         gtok = mean(x["stats"]["generation_tokens"] for x in runs)
                         peak = mean(

bench/run_lcb_tps_bench.sh

@@ -0,0 +1,91 @@
+#!/usr/bin/env bash
+#
+# Run exo_bench.py for each model/mode from bench_params.json.
+#
+# For each entry, runs with:
+#   --pp 800 (fixed, representative LCB prompt length)
+#   --tg <mean completion tokens from vLLM>
+#   --sharding tensor --instance-meta jaccl
+#   --min-nodes 1 --max-nodes 4
+#   --repeat 1
+#   --danger-delete-downloads
+#   --settle-timeout 300
+#
+# Results go to bench/eval_results/<model_dir>/tps_<mode>.json
+#
+# Usage:
+#   bash bench/run_lcb_tps_bench.sh              # run all
+#   bash bench/run_lcb_tps_bench.sh --dry-run    # show what would run
+
+set -euo pipefail
+cd "$(dirname "$0")"
+
+PARAMS_FILE="eval_results/bench_params.json"
+PP=800
+HOST="${EXO_HOST:-s9}"
+DRY_RUN=false
+
+if [[ "${1:-}" == "--dry-run" ]]; then
+    DRY_RUN=true
+fi
+
+if [[ ! -f "$PARAMS_FILE" ]]; then
+    echo "ERROR: $PARAMS_FILE not found. Run compute_bench_params.py first."
+    exit 1
+fi
+
+# Parse bench_params.json and run each entry
+python3 -c "
+import json, sys
+data = json.load(open('$PARAMS_FILE'))
+for entry in data:
+    mlx_id = entry['mlx_model_id']
+    mode = entry['mode']
+    tg = entry['bench_params']['tg']
+    vllm_name = entry['vllm_name']
+    # Output dir: replace / with _
+    out_dir = 'eval_results/' + mlx_id.replace('/', '_')
+    out_file = out_dir + '/tps_' + mode + '.json'
+    print(f'{mlx_id}\t{mode}\t{tg}\t{out_file}\t{vllm_name}')
+" | while IFS=$'\t' read -r model mode tg out_file vllm_name; do
+    out_dir="$(dirname "$out_file")"
+    mkdir -p "$out_dir"
+
+    echo ""
+    echo "============================================================"
+    echo "Model:  $model"
+    echo "Mode:   $mode"
+    echo "vLLM:   $vllm_name"
+    echo "PP:     $PP"
+    echo "TG:     $tg"
+    echo "Output: $out_file"
+    echo "============================================================"
+
+    if [[ -f "$out_file" ]]; then
+        echo "SKIP: $out_file already exists"
+        continue
+    fi
+
+    if [[ "$DRY_RUN" == "true" ]]; then
+        echo "DRY-RUN: would run exo_bench.py"
+        continue
+    fi
+
+    uv run python exo_bench.py \
+        --host "$HOST" \
+        --model "$model" \
+        --pp "$PP" \
+        --tg "$tg" \
+        --repeat 1 \
+        --sharding tensor \
+        --instance-meta jaccl \
+        --min-nodes 1 \
+        --max-nodes 4 \
+        --settle-timeout 300 \
+        --force-download \
+        --danger-delete-downloads \
+        --json-out "$out_file" || echo "FAILED: $model ($mode)"
+done
+
+echo ""
+echo "All benchmarks complete."

dashboard/src/lib/components/ChatForm.svelte

@@ -1,9 +1,6 @@
 <script lang="ts">
   import {
     isLoading,
-    sendMessage,
-    generateImage,
-    editImage,
     editingImage,
     clearEditingImage,
     selectedChatModel,
@@ -28,7 +25,7 @@
     modelTasks?: Record<string, string[]>;
     modelCapabilities?: Record<string, string[]>;
     onSend?: () => void;
-    onAutoSend?: (
+    onAutoSend: (
       content: string,
       files?: {
         id: string;
@@ -216,49 +213,10 @@
     uploadedFiles = [];
     resetTextareaHeight();
 
-    // 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,
-      );
-    }
-
+    // Parent controls all send logic (including image routing,
+    // launching non-running models before sending, etc.)
+    onAutoSend(content, files);
     onSend?.();
-
-    // Refocus the textarea after sending
     setTimeout(() => textareaRef?.focus(), 10);
   }
 

dashboard/src/lib/components/FamilyLogos.svelte

@@ -82,6 +82,12 @@
       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/stores/app.svelte.ts

@@ -1577,6 +1577,7 @@ 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":
@@ -1792,6 +1793,14 @@ class AppStore {
             this.persistConversation(targetConversationId);
           }
         },
+        {
+          generation_stats: (data) => {
+            const stats = data as { generation_tps: number };
+            if (stats.generation_tps > 0) {
+              this.tps = stats.generation_tps;
+            }
+          },
+        },
       );
 
       // Final update
@@ -1989,6 +1998,14 @@ class AppStore {
             this.persistConversation(targetConversationId);
           }
         },
+        {
+          generation_stats: (data) => {
+            const stats = data as { generation_tps: number };
+            if (stats.generation_tps > 0) {
+              this.tps = stats.generation_tps;
+            }
+          },
+        },
       );
 
       // Final cleanup of the message (if conversation still exists)
@@ -2396,7 +2413,7 @@ class AppStore {
 
       let streamedContent = "";
       let streamedThinking = "";
-
+      let serverTpsReceived = false;
       interface ChatCompletionChunk {
         choices?: Array<{
           delta?: { content?: string; reasoning_content?: string };
@@ -2461,7 +2478,6 @@ class AppStore {
             tokenCount += 1;
             this.totalTokens = tokenCount;
 
-            // Update real-time TPS during streaming
             if (firstTokenTime !== null && tokenCount > 1) {
               const elapsed = performance.now() - firstTokenTime;
               this.tps = (tokenCount / elapsed) * 1000;
@@ -2512,16 +2528,24 @@ class AppStore {
               startedAt: this.prefillProgress?.startedAt ?? performance.now(),
             };
           },
+          generation_stats: (data) => {
+            const stats = data as { generation_tps: number };
+
+            if (stats.generation_tps > 0) {
+              this.tps = stats.generation_tps;
+              serverTpsReceived = true;
+            }
+          },
         },
       );
 
       // Clear prefill progress after stream ends
       this.prefillProgress = null;
 
-      // Calculate final TPS
-      if (firstTokenTime !== null && tokenCount > 1) {
+      // Use server-side TPS if available, otherwise fall back to client-side
+      if (!serverTpsReceived && firstTokenTime !== null && tokenCount > 1) {
         const totalGenerationTime = performance.now() - firstTokenTime;
-        this.tps = (tokenCount / totalGenerationTime) * 1000; // tokens per second
+        this.tps = (tokenCount / totalGenerationTime) * 1000;
       }
 
       // Final cleanup of the message (if conversation still exists)

dashboard/src/routes/+page.svelte

@@ -42,6 +42,9 @@
     setSelectedChatModel,
     selectedChatModel,
     sendMessage,
+    generateImage,
+    editImage,
+    editingImage,
     messages,
     debugMode,
     toggleDebugMode,
@@ -834,6 +837,52 @@
     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";
 
@@ -1527,7 +1576,11 @@
           downloadKind
         ] as Record<string, unknown>;
 
-        if (downloadKind !== "DownloadOngoing") continue;
+        if (
+          downloadKind !== "DownloadOngoing" &&
+          downloadKind !== "DownloadPending"
+        )
+          continue;
         if (!downloadPayload) continue;
 
         const downloadModelId = extractModelIdFromDownload(downloadPayload);
@@ -1542,9 +1595,38 @@
           if (downloadModelId !== modelId) continue;
         }
 
-        isDownloading = true;
+        // 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);
+        }
 
-        const progress = parseDownloadProgress(downloadPayload);
         if (progress) {
           // Sum all values across nodes - each node downloads independently
           totalBytes += progress.totalBytes;
@@ -1696,7 +1778,11 @@
           }
         }
 
-        if (downloadKind !== "DownloadOngoing") continue;
+        if (
+          downloadKind !== "DownloadOngoing" &&
+          downloadKind !== "DownloadPending"
+        )
+          continue;
         if (!downloadPayload) continue;
 
         // Check if this download is for this instance's model
@@ -1706,9 +1792,37 @@
           downloadModelId &&
           downloadModelId === instanceModelId
         ) {
-          isDownloading = true;
+          // 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);
+          }
 
-          const progress = parseDownloadProgress(downloadPayload);
           if (progress) {
             // Sum all values across nodes - each node downloads independently
             totalBytes += progress.totalBytes;
@@ -2786,7 +2900,7 @@
         // Running model is same or better tier — use it directly
         setSelectedChatModel(bestRunning.id);
         if (!chatStarted) createConversation();
-        sendMessage(content, files);
+        routeMessage(content, files);
         return;
       }
     }
@@ -2803,7 +2917,7 @@
     if (hasRunningInstance(autoModel.id)) {
       setSelectedChatModel(autoModel.id);
       if (!chatStarted) createConversation();
-      sendMessage(content, files);
+      routeMessage(content, files);
       return;
     }
 
@@ -2956,7 +3070,7 @@
       if (pendingAutoMessage) {
         const msg = pendingAutoMessage;
         pendingAutoMessage = null;
-        sendMessage(msg.content, msg.files);
+        routeMessage(msg.content, msg.files);
       }
       return;
     }
@@ -3035,7 +3149,7 @@
     // Model is selected and running — send directly
     if (model && hasRunningInstance(model)) {
       chatLaunchState = "ready";
-      sendMessage(content, files, null);
+      routeMessage(content, files);
       return;
     }
 

flake.nix

@@ -117,12 +117,13 @@
               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;
+                inherit uvLockMlxVersion uvLockMlxRev;
               };
               default = self'.packages.exo;
             }

nix/mlx.nix

@@ -11,6 +11,7 @@
 , fmt
 , python313Packages
 , uvLockMlxVersion
+, uvLockMlxRev
 }:
 
 assert stdenv.isDarwin;
@@ -41,15 +42,13 @@ let
 
   mlx = stdenv.mkDerivation rec {
     pname = "mlx";
-    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;
+    version = uvLockMlxVersion;
     pyproject = true;
 
     src = fetchFromGitHub {
       owner = "rltakashige";
       repo = "mlx-jaccl-fix-small-recv";
-      rev = "257d5692fc7af6bba3b8afaeb63c549b7d1e43d5";
+      rev = uvLockMlxRev;
       hash = "sha256-GosFIWxIB48Egb1MqJrR3xhsUsQeWdRk5rV93USY6wQ=";
     };
 

pyproject.toml

@@ -25,8 +25,7 @@ dependencies = [
   "openai-harmony>=0.0.8",
   "httpx>=0.28.1",
   "tomlkit>=0.14.0",
-  "pillow>=11.0,<12.0",                        # compatibility with mflux
-  "mflux==0.15.5",
+  "mflux==0.16.9",
   "python-multipart>=0.0.21",
   "msgspec>=0.19.0",
   "zstandard>=0.23.0",
@@ -62,7 +61,7 @@ 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", branch = "main" }
 # 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 }

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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -0,0 +1,12 @@
+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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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

@@ -1,6 +1,7 @@
 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"

resources/inference_model_cards/mlx-community--Qwen3.5-2B-MLX-8bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Qwen3.5-2B-MLX-8bit"
 n_layers = 24
 hidden_size = 2048
+num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-35B-A3B-4bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Qwen3.5-35B-A3B-4bit"
 n_layers = 40
 hidden_size = 2048
+num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-35B-A3B-8bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Qwen3.5-35B-A3B-8bit"
 n_layers = 40
 hidden_size = 2048
+num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-397B-A17B-4bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Qwen3.5-397B-A17B-4bit"
 n_layers = 60
 hidden_size = 4096
+num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-397B-A17B-6bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Qwen3.5-397B-A17B-6bit"
 n_layers = 60
 hidden_size = 4096
+num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-397B-A17B-8bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Qwen3.5-397B-A17B-8bit"
 n_layers = 60
 hidden_size = 4096
+num_key_value_heads = 2
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"

resources/inference_model_cards/mlx-community--Qwen3.5-9B-4bit.toml

@@ -1,6 +1,7 @@
 model_id = "mlx-community/Qwen3.5-9B-4bit"
 n_layers = 32
 hidden_size = 4096
+num_key_value_heads = 4
 supports_tensor = true
 tasks = ["TextGeneration"]
 family = "qwen"