Zhihao Wang 72696ef1c7 [fix] DeepSeek-V4 blockwise FP8: keep linear_weights_proj to bf16 (#1505) ...

Signed-off-by: zhihaow6 <zhihaow6@illinois.edu>

2026-06-29 12:56:47 -07:00

..

visualize

[tool] CPU mem: add CPU profiler and visualizer (#717)

2026-03-09 21:55:19 -07:00

__init__.py

[tool] CPU mem: add CPU profiler and visualizer (#717)

2026-03-09 21:55:19 -07:00

convert_fsdp_to_hf.py

[refactor] hf_config: single entry point with alias registration and overrides (#1216)

2026-05-28 13:14:26 -07:00

convert_hf_to_fp8.py

[fix] DeepSeek-V4 blockwise FP8: keep linear_weights_proj to bf16 (#1505)

2026-06-29 12:56:47 -07:00

convert_hf_to_hf_int4.py

[refactor] hf_tokenizer: route AutoTokenizer call sites through load_tokenizer (#1237)

2026-05-28 13:33:18 -07:00

convert_hf_to_int4_direct.py

2/5 support kimi 2.5 full + lora: VL-aware quantization/conversion tools (#1220)

2026-05-28 14:46:27 -07:00

convert_hf_to_int4.py

2/5 support kimi 2.5 full + lora: VL-aware quantization/conversion tools (#1220)

2026-05-28 14:46:27 -07:00

convert_hf_to_mxfp8.py

[feat] Expand mxfp8 quantization to support fine-grained layer precision config (#614)

2026-04-21 13:10:44 -07:00

convert_hf_to_nvfp4.py

Use TE NVFP4 reference quantizer and add bitwise exact tests (#1054)

2026-05-01 00:19:10 -07:00

convert_hf_to_torch_dist.py

deepseek v4 model support (#1181)

2026-06-04 20:12:04 -07:00

convert_kimi_int4_to_bf16.py

2/5 support kimi 2.5 full + lora: VL-aware quantization/conversion tools (#1220)

2026-05-28 14:46:27 -07:00

convert_to_hf.py

[refactor] hf_tokenizer: route AutoTokenizer call sites through load_tokenizer (#1237)

2026-05-28 13:33:18 -07:00

convert_torch_dist_to_hf_ray.py

fix(precommit): route load_hf_config through miles helper (#1484)

2026-06-25 14:58:52 -07:00

convert_torch_dist_to_hf.py

[refactor] hf_config: single entry point with alias registration and overrides (#1216)

2026-05-28 13:14:26 -07:00

cpu_memory_profiler.py

[tool] CPU mem: add CPU profiler and visualizer (#717)

2026-03-09 21:55:19 -07:00

fp8_cast_bf16.py

deepseek v4 model support (#1181)

2026-06-04 20:12:04 -07:00

param_name_remap.py

deepseek v4 model support (#1181)

2026-06-04 20:12:04 -07:00