31.5 GB

Ctrl+K

1 contributor

Calibrated FP8 KV-cache scales (mla_kv_cache_scales); FP8 KV by default; card: family, FP8 KV, MTP caveat. Needs qwen-mla-vllm >= 0.5.0

e9bd978 verified about 1 month ago

.gitattributes

1.57 kB
Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
README.md

9.86 kB
Calibrated FP8 KV-cache scales (mla_kv_cache_scales); FP8 KV by default; card: family, FP8 KV, MTP caveat. Needs qwen-mla-vllm >= 0.5.0 about 1 month ago
chat_template.jinja

8.95 kB
Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
config.json

7.59 kB
Calibrated FP8 KV-cache scales (mla_kv_cache_scales); FP8 KV by default; card: family, FP8 KV, MTP caveat. Needs qwen-mla-vllm >= 0.5.0 about 1 month ago
generation_config.json

202 Bytes
Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
merges.txt

3.35 MB
Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-00001-of-00007.safetensors

2.54 GB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-00002-of-00007.safetensors

4.97 GB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-00003-of-00007.safetensors

4.97 GB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-00004-of-00007.safetensors

4.99 GB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-00005-of-00007.safetensors

4.99 GB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-00006-of-00007.safetensors

4.95 GB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-00007-of-00007.safetensors

2.31 GB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
model-mtp.safetensors

849 MB
xet

Add the base model's vision tower and MTP head (Qwen3_5MLAForConditionalGeneration); card: vision + MTP results. Needs qwen-mla-vllm >= 0.4.0 about 1 month ago
model-visual.safetensors

921 MB
xet

Add the base model's vision tower and MTP head (Qwen3_5MLAForConditionalGeneration); card: vision + MTP results. Needs qwen-mla-vllm >= 0.4.0 about 1 month ago
model.safetensors.index.json

151 kB
Add the base model's vision tower and MTP head (Qwen3_5MLAForConditionalGeneration); card: vision + MTP results. Needs qwen-mla-vllm >= 0.4.0 about 1 month ago
preprocessor_config.json

390 Bytes
Add the base model's vision tower and MTP head (Qwen3_5MLAForConditionalGeneration); card: vision + MTP results. Needs qwen-mla-vllm >= 0.4.0 about 1 month ago
tokenizer.json

12.8 MB
xet

Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
tokenizer_config.json

17.9 kB
Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago
video_preprocessor_config.json

385 Bytes
Add the base model's vision tower and MTP head (Qwen3_5MLAForConditionalGeneration); card: vision + MTP results. Needs qwen-mla-vllm >= 0.4.0 about 1 month ago
vocab.json

6.72 MB
Qwen3.8-27B-MLA-FP8: FP8 block build of the training-free retrofit about 1 month ago