Yueming Yuan 6d0cd93f10 DeepSeek-V4 support-v2: native megatron implementation as option (#2706)

2026-08-31 13:21:05 -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

Use TransformerEngine for MXFP8 quantization (#1935)

2026-07-29 12:39:54 -07:00

convert_hf_to_nvfp4.py

Compact NVFP4 BF16 MoE exclusion metadata (#2043)

2026-08-03 14:48:04 -07:00

convert_hf_to_torch_dist.py

DeepSeek-V4 support-v2: native megatron implementation as option (#2706)

2026-08-31 13:21:05 -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_mxfp4_to_fp8.py

add DeepSeek-V4-Flash-0731 support and mxfp4->fp8 converter (#2717)

2026-08-23 23:42:36 -07:00

convert_to_hf.py

Move FT modules into topic folders (mechanical) (#1596)

2026-07-10 11:44:42 +08:00

convert_torch_dist_to_hf_ray.py

Fix typo environment variable and unbuffer python outputs (#1895)

2026-08-09 18:19:56 +08:00

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

add DeepSeek-V4-Flash-0731 support and mxfp4->fp8 converter (#2717)

2026-08-23 23:42:36 -07:00

param_name_remap.py

deepseek v4 model support (#1181)

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