Commit Graph

  • 60a11a6410 Use Latest Cut Cross Entropy (#2392) xzuyn 2025-03-10 05:26:40 -04:00
  • 46a045e528 chore(doc): add faq when having no default chat_template (#2398) NanoCode012 2025-03-10 16:25:50 +07:00
  • 3b477e08a0 feat(doc): add more info on RewardModel datasets (#2391) NanoCode012 2025-03-10 16:25:31 +07:00
  • 754817c8c6 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-07 13:59:52 +00:00
  • 48846db8e3 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-07 13:59:21 +00:00
  • 16dc6ee68d refactor: trl grpo configs to have descriptions (#2386) NanoCode012 2025-03-07 20:58:53 +07:00
  • fa7c79b3b9 remove lion-pytorch as it's already handled upstream (#2389) Wing Lian 2025-03-07 08:58:15 -05:00
  • d8eb8703f3 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-06 16:50:20 +00:00
  • b6083ae7b1 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-06 16:49:40 +00:00
  • ae66374156 Optimizer refactor and add Muon support (#2367) Wing Lian 2025-03-06 11:49:19 -05:00
  • 5e21b1a9da various fixes 20250305 (#2384) Wing Lian 2025-03-06 11:48:44 -05:00
  • 76bb09784d fix import optimizers-refactor Wing Lian 2025-03-05 14:05:27 -05:00
  • 92f0ed2e37 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-05 16:16:51 +00:00
  • 575e5f28ec Update Tokenizer Overrides Handling in models.py (#1549) mhenrichsen 2025-03-05 17:15:12 +01:00
  • 0542c7dd56 add muon optimizer Wing Lian 2025-02-27 23:15:41 -05:00
  • 3ee96165d3 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-05 15:27:11 +00:00
  • 0134093acc Add REX LR Scheduler (#2380) xzuyn 2025-03-05 10:26:11 -05:00
  • e6453035a7 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-05 15:03:12 +00:00
  • d4de93a7bb feat(grpo): add reward_weights config and refactor (#2365) NanoCode012 2025-03-05 22:02:08 +07:00
  • c8191394e9 fix(doc): add missing low_cpu_mem_usage config to docs (#2369) [skip ci] NanoCode012 2025-03-05 22:01:44 +07:00
  • f18231c653 chore(doc): add clarification about mpi4py error on single gpu deepspeed (#2383) [skip ci] NanoCode012 2025-03-05 22:01:28 +07:00
  • 9ed4f6b3aa feat(doc): document drop_system_message and clarify limitation (#2381) [skip ci] NanoCode012 2025-03-05 22:01:16 +07:00
  • 05dddfc41d feat(doc): add docker images explanation (#2379) [skip ci] NanoCode012 2025-03-05 22:01:00 +07:00
  • 8e30917440 chore(docs): remove phorm (#2378) [skip ci] NanoCode012 2025-03-05 22:00:50 +07:00
  • d883b11b6f fix(doc): add installation for cce to docs (#2375) [skip ci] NanoCode012 2025-03-05 22:00:39 +07:00
  • 9270933a28 Built site for gh-pages Quarto GHA Workflow Runner 2025-03-05 13:59:40 +00:00
  • f4910dd2ea train.py refactor (#2371) Dan Saunders 2025-03-05 08:58:33 -05:00
  • 5c0510a876 review comments train-refactor Dan Saunders 2025-03-03 18:44:16 +00:00
  • e1bc18763a combine like functions Dan Saunders 2025-02-28 17:47:39 +00:00
  • 10d18e6c97 fix(test): replace jackfram llama with smollm fix/replace_jackllama NanoCode012 2025-02-28 16:40:49 +07:00
  • ee489d16bf wip seq-parallel-ring Wing Lian 2025-02-27 11:42:46 -05:00
  • ed5178cd3d update Dan Saunders 2025-02-26 21:03:44 +00:00
  • a3224c7c3c updates Dan Saunders 2025-02-26 20:31:54 +00:00
  • c4104fc10c refactor train.py Dan Saunders 2025-02-26 19:37:42 +00:00
  • 68e97d032a chunk to prevent overflows in kernel topk-logprobs-triton Wing Lian 2025-02-26 04:44:24 -05:00
  • 23f029a89c lint and additional train metric checks for kd Wing Lian 2025-02-26 03:19:42 -05:00
  • afbb44f08b more optims Wing Lian 2025-02-26 01:49:47 -05:00
  • d753ead033 optimize and include bench util Wing Lian 2025-02-26 01:17:50 -05:00
  • f68aedd1f8 Update __init__.py tp_support Sung Ching Liu 2025-02-21 11:13:15 -05:00
  • 3dd5c6f8ec nit Sunny Liu 2025-02-21 10:59:03 -05:00
  • 4caa59a087 auto detect tp_size Sunny Liu 2025-02-20 11:36:55 -05:00
  • 984be14147 add tp_size in config doc Sunny Liu 2025-02-20 00:01:59 -05:00
  • 64adbf1a15 tp plan not needed Sunny Liu 2025-02-19 19:18:36 -05:00
  • 438b623031 prepare accelerate envs for tp Sunny Liu 2025-02-18 23:34:05 -05:00
  • a74efcecbe skip move to device Sunny Liu 2025-02-18 17:13:52 -05:00
  • d663652216 del device_map for tp Sunny Liu 2025-02-18 17:08:32 -05:00
  • dbd43aa18f set tp_plan Sunny Liu 2025-02-18 17:04:13 -05:00
  • dbdf97e828 enabe tp thru tp_size Sunny Liu 2025-02-18 16:20:24 -05:00
  • c011405117 fix gradients Wing Lian 2025-02-25 23:34:27 -05:00
  • f144319697 distributed fix telemetry Dan Saunders 2025-02-26 02:55:44 +00:00
  • 07bb41812b fix issue with tests in ci Dan Saunders 2025-02-24 21:30:34 +00:00
  • cae8c7636b fixes Dan Saunders 2025-02-24 20:30:16 +00:00
  • 09611cea10 remove duplicate info Dan Saunders 2025-02-24 20:02:16 +00:00
  • 61266ab843 adding runtime metrics / system info additional accelerator support, etc. Dan Saunders 2025-02-24 19:37:11 +00:00
  • aea0e760e4 adding runtime metrics / system info additional accelerator support, etc. Dan Saunders 2025-02-24 19:36:31 +00:00
  • 5afad670da improved redaction, send system info during model config load telemetry, etc. Dan Saunders 2025-02-24 15:39:02 +00:00
  • 49ac79ed1e doc update Dan Saunders 2025-02-24 01:49:31 +00:00
  • c9af72cd7a fix Dan Saunders 2025-02-24 01:31:35 +00:00
  • d3d63c1432 adding back in base_model redaction w/ whitelist Dan Saunders 2025-02-24 01:16:03 +00:00
  • 675b65d711 sleep on all ranks in distributed setting Dan Saunders 2025-02-24 00:53:58 +00:00
  • b23187daea simplifying path redaction Dan Saunders 2025-02-24 00:06:08 +00:00
  • e373a6b8d0 small update / fix Dan Saunders 2025-02-21 20:35:09 +00:00
  • fd5d5aecdc tests for runtime metrics telemetry and assoc. callback Dan Saunders 2025-02-21 20:31:07 +00:00
  • 3760175440 adding runtime metrics (cpu + gpu memory, steps/s, etc.) Dan Saunders 2025-02-21 19:01:35 +00:00
  • 7927abff90 updated sanitization logic, tests Dan Saunders 2025-02-24 20:05:55 +00:00
  • ec36839316 update error file path sanitization function; adding more error tracking Dan Saunders 2025-02-21 13:57:08 +00:00
  • 3076b8df00 progress on telemetry: config load, process, model load, train start / end, error tracking Dan Saunders 2025-02-19 22:05:12 +00:00
  • c50610375f updates Dan Saunders 2025-02-19 13:55:04 +00:00
  • 07ffd47f2b updates Dan Saunders 2025-02-19 13:54:51 +00:00
  • 76d951afd2 adding todo Dan Saunders 2025-02-17 22:25:54 +00:00
  • 5220e8ccf4 initial telemetry manager impl Dan Saunders 2025-02-17 18:31:42 +00:00
  • 1a3f9bcbbe Built site for gh-pages Quarto GHA Workflow Runner 2025-02-25 18:51:14 +00:00
  • 75cbd15301 Fix(doc): address missing doc changes (#2362) v0.7.1 NanoCode012 2025-02-26 01:50:02 +07:00
  • a2e52a29e9 chore: lint Wing Lian 2025-02-25 07:29:46 -05:00
  • efdea0e9ff Built site for gh-pages Quarto GHA Workflow Runner 2025-02-25 09:11:03 +00:00
  • 2efe1b4c09 Feat(doc): Reorganize documentation, fix broken syntax, update notes (#2348) NanoCode012 2025-02-25 16:09:37 +07:00
  • e82268e580 use triton for kd-loss in trainer Wing Lian 2025-02-24 22:58:35 -05:00
  • 75e1480c10 chunking not necessary Wing Lian 2025-02-24 22:56:15 -05:00
  • 45e1548d59 fix the kernels Wing Lian 2025-02-24 22:38:55 -05:00
  • 165088e7c1 triton kernel for top-k logprob kd Wing Lian 2025-02-24 22:13:26 -05:00
  • 4c1ebd7402 Built site for gh-pages Quarto GHA Workflow Runner 2025-02-24 20:04:15 +00:00
  • 1110a37e21 feat: add deepseek_v3 sample packing (#2230) NanoCode012 2025-02-25 03:03:15 +07:00
  • 7f31599da4 Built site for gh-pages Quarto GHA Workflow Runner 2025-02-24 17:41:49 +00:00
  • 9850f42204 bump liger to 0.5.3 (#2353) Wing Lian 2025-02-24 12:40:54 -05:00
  • f897224306 Built site for gh-pages Quarto GHA Workflow Runner 2025-02-24 16:13:58 +00:00
  • 00fc8109e4 Correctly reference mount paths (#2347) Matt Baker 2025-02-24 08:12:57 -08:00
  • e618244a70 Built site for gh-pages Quarto GHA Workflow Runner 2025-02-23 17:32:31 +00:00
  • 2d5826f544 Relicense the logprob KD loss functions as Apache 2.0 (#2358) Wing Lian 2025-02-23 12:31:35 -05:00
  • d88e071120 sequence parallelism Wing Lian 2025-02-23 12:19:34 -05:00
  • 17fb239c63 Built site for gh-pages Quarto GHA Workflow Runner 2025-02-22 03:07:26 +00:00
  • a4170030ab don't install extraneous old version of pydantic in ci and make sre to run multigpu ci (#2355) Wing Lian 2025-02-21 22:06:29 -05:00
  • 328bb0466b Merge branch 'main' into flx_attn_support flx_attn_support Sung Ching Liu 2025-02-21 11:27:25 -05:00
  • e792b54bab remove unnecessary components Sunny Liu 2025-02-21 11:23:21 -05:00
  • bf842730a5 fix(doc): add missing auto_find_batch_size (#2339) [skip ci] NanoCode012 2025-02-21 11:56:38 +07:00
  • 248a0b9f6f Built site for gh-pages Quarto GHA Workflow Runner 2025-02-21 03:57:38 +00:00
  • 3d80a2df6d Built site for gh-pages Quarto GHA Workflow Runner 2025-02-21 03:57:03 +00:00
  • 1db6ad60a7 support for passing init_lora_weights to lora_config (#2352) Wing Lian 2025-02-20 22:56:34 -05:00
  • 29b366b2e1 Bumping 0.15.1 TRL version for GRPO+PEFT fix (#2344) salman 2025-02-21 03:56:04 +00:00
  • a9ebff087c remove ref_model when peft model is passed into grpo trainer grpo-ref-model-cleanup Wing Lian 2025-02-20 21:53:20 -05:00
  • 10518d9ec0 Built site for gh-pages Quarto GHA Workflow Runner 2025-02-21 02:13:04 +00:00