mlx-lm

__init__.py

Mlx llm package (#301 )

2024-01-12 10:25:56 -08:00

activations.py

move Xielu Activation in Apertus to activations.py (#772 )

2026-01-20 06:51:16 -08:00

afm7.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

afmoe.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

apertus.py

Apertus tie_word_embeddings fix (#1143 )

2026-04-20 23:47:40 +02:00

baichuan_m1.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

bailing_moe_linear.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

bailing_moe.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

base.py

Batch support for mamba-style models (#468 )

2025-09-16 08:01:45 -07:00

bitlinear_layers.py

Feat: add falcon-e support for bitnet models (#268 )

2025-07-07 08:43:52 -07:00

bitnet.py

Update bitnet, nemotron h to use build in relu2 from MLX (#446 )

2025-09-13 07:02:29 -07:00

cache.py

Fix ArraysCache extend (#1177 )

2026-04-21 16:41:49 -07:00

cohere2.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

cohere.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

dbrx.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

deepseek_v2.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

deepseek_v3.py

DSV3 MLA (#839 )

2026-02-04 12:06:42 -08:00

deepseek_v32.py

Batch generation refactoring and various fixes (#1072 )

2026-04-01 15:07:50 -07:00

deepseek.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

dots1.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

ernie4_5_moe.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

ernie4_5.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

exaone4.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

exaone_moe.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

exaone.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

falcon_h1.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

gated_delta.py

Batch generation refactoring and various fixes (#1072 )

2026-04-01 15:07:50 -07:00

gemma2.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

gemma3_text.py

fix dsv32 and gemma3 (#664 )

2025-12-08 16:14:10 -08:00

gemma3.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

gemma3n.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

gemma4_text.py

Fix Gemma 4 sanitize() not stripping KV projections for shared layers (#1240 )

2026-05-04 15:26:18 -07:00

gemma4.py

Add gemma 4 (#1093 )

2026-04-04 04:47:03 -07:00

gemma.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

glm4_moe_lite.py

DSV3 MLA (#839 )

2026-02-04 12:06:42 -08:00

glm4_moe.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

glm4.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

glm_moe_dsa.py

Add GLM5 (#867 )

2026-02-12 10:11:16 -08:00

glm.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

gpt2.py

GPT2 Batching Fix (#529 )

2025-10-07 00:36:48 -07:00

gpt_bigcode.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

gpt_neox.py

add parallel_residual setting to gptneox (#586 )

2025-11-05 07:52:08 -08:00

gpt_oss.py

Add gpt-oss sharding (#761 )

2026-01-17 07:42:11 -08:00

granite.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

granitemoe.py

model: GraniteMoeHybrid (#442 )

2025-09-10 08:37:32 -07:00

granitemoehybrid.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

helium.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

hunyuan_v1_dense.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

hunyuan.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

internlm2.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

internlm3.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

iquestloopcoder.py

Fix batch generation for IQuestLoopCoder model (#748 )

2026-01-09 16:50:51 -08:00

jamba.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

kimi_k25.py

Add Kimi-K2.5 (#813 )

2026-01-27 10:23:39 -08:00

kimi_linear.py

fix: break shared-buffer memory leak in GatedDeltaNet cache (#1077 )

2026-03-31 21:27:11 -07:00

kimi_vl.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

Klear.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

lfm2_moe.py

Allow reading LFM2 models nested rope params (#908 )

2026-02-18 16:25:54 -08:00

lfm2-vl.py

fix for LFM2 (#493 )

2025-09-23 13:12:28 -07:00

lfm2.py

Allow reading LFM2 models nested rope params (#908 )

2026-02-18 16:25:54 -08:00

lille-130m.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

llama4_text.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

llama4.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

llama.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

longcat_flash_ngram.py

LongCat MLA (#868 )

2026-02-12 06:54:44 -08:00

longcat_flash.py

LongCat MLA (#868 )

2026-02-12 06:54:44 -08:00

mamba2.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

mamba.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

mimo_v2_flash.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

mimo.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

minicpm3.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

minicpm.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

minimax.py

Fix sharded rms norm in MiniMax M2.5 (#898 )

2026-02-16 17:20:07 -08:00

ministral3.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

mistral3.py

Ministral3 (#642 )

2025-12-02 10:59:45 -08:00

mixtral.py

Add tie_word_embeddings modulars in mistral and qwen3 moe (#889 )

2026-02-18 11:23:40 -08:00

mla.py

Fix mixed quant predicates for MLA models (#892 )

2026-02-15 02:44:01 -08:00

nanochat.py

Add support for nanochat (#554 )

2025-10-15 12:59:54 -07:00

nemotron_h.py

Fix SSM dt clamp default for Nemotron-H (#1026 )

2026-03-30 04:19:27 -07:00

nemotron-nas.py

Adds support for Nemotron Super 49b v1.5 (#756 )

2026-01-16 07:03:05 -08:00

nemotron.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

olmo2.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

olmo3.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

olmo.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

olmoe.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

openelm.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

phi3.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

phi3small.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

phi.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

phimoe.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

phixtral.py

Length masking for batch inputs (#1173 )

2024-12-18 19:43:52 -08:00

pipeline.py

DWQ for very large models (#536 )

2025-11-07 06:43:40 -08:00

pixtral.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

plamo2.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

plamo.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

qwen2_moe.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

qwen2_vl.py

fix loading for qwen2 VL (#491 )

2025-09-23 13:12:37 -07:00

qwen2.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

qwen3_5_moe.py

fix qwen3.5 sanitize (#928 )

2026-02-24 17:04:43 -08:00

qwen3_5.py

Batch generation refactoring and various fixes (#1072 )

2026-04-01 15:07:50 -07:00

qwen3_moe.py

Add tie_word_embeddings modulars in mistral and qwen3 moe (#889 )

2026-02-18 11:23:40 -08:00

qwen3_next.py

fix: break shared-buffer memory leak in GatedDeltaNet cache (#1077 )

2026-03-31 21:27:11 -07:00

qwen3_vl_moe.py

Add Qwen3-VL language model implementation (#547 )

2025-10-09 21:42:46 -07:00

qwen3_vl.py

Add Qwen3-VL (Dense) language model implementation (#553 )

2025-10-14 12:28:46 -07:00

qwen3.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

qwen.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

recurrent_gemma.py

Fix batch mamba (#842 )

2026-02-03 19:31:42 -08:00

rope_utils.py

Add gemma 4 (#1093 )

2026-04-04 04:47:03 -07:00

rwkv7.py

Add RWKV7 (#580 )

2025-12-31 14:21:42 -08:00

seed_oss.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

smollm3.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

solar_open.py

Add Solar Open (#721 )

2026-01-05 08:38:53 -08:00

ssm.py

Nemotron super support (#992 )

2026-03-16 10:59:14 -07:00

stablelm.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00

starcoder2.py

some cleanup + tests towards batching (#430 )

2025-09-09 13:05:11 -07:00

step3p5.py

step3p5: use rotating cache for sliding attention layers (#949 )

2026-03-04 17:17:29 -08:00

switch_layers.py

Fix swiglu parameter order (#767 )

2026-01-17 06:13:23 -08:00

telechat3.py

Adding TeleChat3 (#773 )

2026-01-22 07:35:58 -08:00

youtu_llm.py

Use compiled Swiglu (#753 )

2026-01-14 11:58:41 -08:00