github
/
minimind
mirror of https://github.com/jingyaogong/minimind.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Commit Graph

  • 6fc918beb6 [update] remove expert args jingyaogong 2026-09-04 22:16:38 +0800
  • 4bd6062ead
         Merge pull request #836 from JustinGuese/moe-cli-flags jingyaogong 2026-09-04 22:10:18 +0800
  • 4d7a74acab
         Merge pull request #840 from JustinGuese/readme-benchmark-resolution jingyaogong 2026-09-04 22:10:00 +0800
  • 77946fda00 [docs] 在客观评测表下补充各数据集的测量分辨率 Justin Guese 2026-09-03 17:10:52 +0700
  • fc585a9bf9 [update] 把随机种子暴露到命令行 Justin Guese 2026-09-03 15:36:14 +0700
  • c3199c7240 [update] 把 num_experts / num_experts_per_tok 暴露到命令行 Justin Guese 2026-09-03 15:13:29 +0700
  • 6b2f4c939a [update] MoE 分发：每层主机同步从 3E 次降到 1 次 Justin Guese 2026-09-03 15:05:25 +0700
  • 56a21ed3c2 docs: expand bilingual dataset guide AmirX0 2026-09-01 07:06:36 +0330
  • 5b26153b75 fix: make repetition penalty multilingual AmirX0 2026-09-01 07:03:34 +0330
  • 66aef88185 docs: add multilingual README entry points AmirX0 2026-09-01 06:55:44 +0330
  • e556464734
         Merge efd8073450 into 7a6fddd63a yanghw116 2026-09-01 10:09:16 +0800
  • 7a6fddd63a [update] readme jingyaogong 2026-08-31 18:08:05 +0800
  • c9a8ab4632 [update] dependency version jingyaogong 2026-08-31 17:54:37 +0800
  • 56d4cddfd8 [fix] keep DPO preference pair think formatting aligned Daoyuan Li 2026-08-29 18:29:00 -0700
  • e84156818a [update] dependency version jingyaogong 2026-08-29 15:12:28 +0800
  • 21b8a8b70a feat: 添加 git 交互脚本 jinhu 2026-08-27 22:30:52 +0800
  • 2b250aca51 chore: remove .idea and add to .gitignore jinhu 2026-08-27 22:10:02 +0800
  • 0c91dea4a2 [refactor] remove redundant rollout update_policy after torch.compile TOPSAIL\78190 2026-08-24 12:43:50 +0800
  • 387925b6bc [update] readme content jingyaogong 2026-08-21 16:49:00 +0800
  • d65ef2c00e [update] readme content jingyaogong 2026-08-21 16:49:00 +0800
  • 8425924da2 [update] readme content jingyaogong 2026-08-21 16:30:30 +0800
  • ed5ea9e02a [update] readme content jingyaogong 2026-08-21 16:30:30 +0800
  • 32928c3a33 [update] readme content jingyaogong 2026-08-21 16:25:20 +0800
  • eabad706da [update] readme content jingyaogong 2026-08-21 16:25:20 +0800
  • 437eae89d6 [feat] add MPS support for Apple Silicon Jiyang Liu 2026-08-14 01:26:49 +0800
  • 1a18e651f6 Add CE regularization term to DPO loss QZL-max 2026-08-12 22:53:12 +0800
  • f7ea0bc4f6 [update] supporter charts jingyaogong 2026-08-06 22:07:29 +0800
  • 393e387e9a [update] supporter charts jingyaogong 2026-08-06 22:07:29 +0800
  • 100a49bad2 feat: make code agent runs reproducible RileyChen23 2026-08-01 16:18:41 +0800
  • 73fb393f00 [docs] add beginner-friendly learning tutorial (docs/) emmmdty 2026-07-31 09:15:01 +0800
  • f07588a31b [refactor] PPO cleanup jingyaogong 2026-07-23 11:11:05 +0800
  • 89d674b8a5 [refactor] PPO cleanup jingyaogong 2026-07-23 11:11:05 +0800
  • 185baaa022 [fix] PPO rollout mode jingyaogong 2026-07-23 11:05:51 +0800
  • 2d4c385b0f [fix] PPO rollout mode jingyaogong 2026-07-23 11:05:51 +0800
  • 1b43526b16 Merge pull request #814 from FloatFrank/fix/ppo-logp-autocast-align-v3 jingyaogong 2026-07-23 10:32:06 +0800
  • e9f46506da
         Merge pull request #814 from FloatFrank/fix/ppo-logp-autocast-align-v3 jingyaogong 2026-07-23 10:32:06 +0800
  • 27b9c0082f [fix] PPO: align mb_resp_logp with old_resp_logp under autocast FloatFrank 2026-07-21 18:51:42 +0800
  • 1191fc8e9f [fix] PPO: align mb_resp_logp with old_resp_logp under autocast FloatFrank 2026-07-21 18:51:42 +0800
  • fbd0b49bb1 [fix] PPO: align mb_resp_logp with old_resp_logp under autocast FloatFrank 2026-07-21 18:08:25 +0800
  • b9aaaabd66 [fix] PPO: align mb_resp_logp with old_resp_logp under autocast FloatFrank 2026-07-21 17:49:53 +0800
  • 864bd67437 feat: add generalized on-policy distillation trainer liuhy495 2026-07-15 21:15:10 +0800
  • 1e0981d1e3 feat: make code generation runs resumable RileyChen23 2026-07-14 20:27:11 +0800
  • cf2329fab0 feat: add reproducible MBPP code evaluation RileyChen23 2026-07-14 18:42:22 +0800
  • 8cbbda005c fix: tighten runtime and GRPO metrics RileyChen23 2026-07-14 18:41:40 +0800
  • 6a98206aef docs: keep repository documentation project-only RileyChen23 2026-07-13 19:19:38 +0800
  • 04f5fd0e8c
         Merge 855c3804ce into 512eed0b65 OrbisAI Security 2026-07-13 08:46:24 +0530
  • 855c3804ce fix: V-002 security vulnerability orbisai0security 2026-07-13 03:15:17 +0000
  • 4ec1b84d96 docs: document code agent and neutralize identity RileyChen23 2026-07-12 17:56:29 +0800
  • 65def67cee feat: add code agent model service runner RileyChen23 2026-07-12 17:53:02 +0800
  • 125b49dfc1 feat: integrate verifiable rewards into GRPO RileyChen23 2026-07-12 17:47:36 +0800
  • bfb5a1261b feat: add execution-feedback repair agent RileyChen23 2026-07-12 17:42:52 +0800
  • 07643d5f05 feat: add pass-at-k code evaluation RileyChen23 2026-07-12 17:38:20 +0800
  • 8e06fb6d52 feat: add code execution verifier and rewards RileyChen23 2026-07-12 17:29:07 +0800
  • d5bee32c6a
         Merge 840cf8a6c6 into 512eed0b65 OrbisAI Security 2026-07-10 07:56:55 +0530
  • 840cf8a6c6 fix: V-001 security vulnerability orbisai0security 2026-07-10 02:25:44 +0000
  • 450885c0d5 feat: 添加 macOS Metal (MPS) 设备支持 shiguoliang 2026-05-24 10:52:56 +0800
  • b40eb54434 FSDP TP Longfang Zhao 2026-07-01 12:02:28 -0700
  • da93141a0d update ignore Longfang Zhao 2026-07-01 11:32:20 -0700
  • 21d501f385 profile. Longfang Zhao 2026-07-01 10:53:44 -0700
  • 4751e359da distributed training. Longfang Zhao 2026-07-01 00:17:57 -0700
  • 02b61435b1 fix Longfang Zhao 2026-06-30 23:23:21 -0700
  • a6a19aa877 update Longfang Zhao 2026-06-30 21:36:03 -0700
  • b59f54cfc5 Add script to download skypile data Longfang Zhao 2026-06-30 14:13:37 -0700
  • 2ce02ea4ab optimize SFT training speed Longfang Zhao 2026-06-30 10:55:27 -0700
  • 17f67f9897 Add MFU cal utils. Longfang Zhao 2026-06-30 09:30:04 -0700
  • c1d990c81c 提升训练速度 Longfang Zhao 2026-06-29 22:05:06 -0700
  • 26e83b0792 add ignore Longfang Zhao 2026-06-29 21:26:21 -0700
  • 63a1bcc82f Add qat Longfang Zhao 2026-06-29 17:40:09 -0700
  • 237f25bc70 [fix] robustness jingyaogong 2026-06-28 18:03:34 +0800
  • 512eed0b65 [fix] robustness jingyaogong 2026-06-28 18:03:34 +0800
  • 857bcbc4f2 Add course entrypoint Sun Yuhao 2026-06-28 14:26:02 +0800
  • 53a4a8fc63 Add MiniMind source course materials Sun Yuhao 2026-06-28 14:00:16 +0800
  • 3108ee24b5 fix: don't char-slice the prompt by max_new_tokens / max_tokens swaylq 2026-06-17 04:43:13 +0800
  • 3233d9872c agentic baihaochen 2026-06-12 15:11:31 +0800
  • 5012b232ff fix: harden DeepSpeed SFT resume baihaochen 2026-06-11 21:45:31 +0800
  • eaf3a8cbec feat: add pretrain evaluation scripts baihaochen 2026-06-07 19:46:48 +0800
  • 4dcda45bfa fix: recover DeepSpeed resume step from checkpoint tag baihaochen 2026-06-04 14:20:21 +0800
  • e9c93951cc fix: keep only latest DeepSpeed checkpoint baihaochen 2026-06-03 21:35:47 +0800
  • 0425c4f622 fix: make DeepSpeed resume checkpoints robust baihaochen 2026-06-02 20:41:38 +0800
  • d97e47c408 day01: 学习笔记 + 调试配置 + batch_size 默认值调整 QiaohuiGao 2026-06-02 00:51:12 -0400
  • b23a05868d Add Day 1 learning notes and model Chinese comments QiaohuiGao 2026-06-01 21:33:23 -0400
  • 33233fed6e feat: add SearchShortQA DeepSpeed pipeline baihaochen 2026-06-02 00:59:46 +0800
  • 13739bdbc5 [fix] ddp exit hang jingyaogong 2026-06-01 17:50:39 +0800
  • 3f1a7cc25b [fix] ddp exit hang jingyaogong 2026-06-01 17:50:39 +0800
  • ccb352f901 docs: add windows tips in README_EN.md Dujus 2026-06-01 11:13:08 +0800
  • 99c26e9043 docs: optimize installation steps for Windows beginners and add HF mirror tip Dujus 2026-06-01 10:59:18 +0800
  • e1df1a50d7 [fix] lora dims jingyaogong 2026-05-31 13:38:51 +0800
  • 4a68da72d5 [fix] lora dims jingyaogong 2026-05-31 13:38:51 +0800
  • b92396ad0a use wandb R 2026-05-28 22:09:21 -0700
  • 0761ffa7cd feat(trainer): add new train_agentic_rl_async.py 王翛玄 2026-05-28 17:01:00 +0800
  • 3033b162c3
         Merge 01c04f519b into 4497610ec0 guosj 2026-05-23 17:19:25 +0800
  • 2201060169 fix: switch DeepSpeed config from bf16 to fp16, ZeRO-2 to ZeRO-1 (RTX 3090 compat) baihaochen 2026-05-23 16:45:56 +0800
  • 46dabcfc19 chore: remove unused import baihaochen 2026-05-23 16:42:02 +0800
  • 907f31f357 chore: add deepspeed suffix to save_weight, update default data_path baihaochen 2026-05-23 16:38:49 +0800
  • 4f5cdfe59a fix: avoid double init_process_group in DeepSpeed pretrain baihaochen 2026-05-23 16:21:57 +0800
  • 6d974260cb
         Merge b7f758a87a into 4497610ec0 Blossom 2026-05-23 15:26:37 +0800
  • b7f758a87a docs: improve README framing for sharing and onboarding MackDing 2026-05-23 15:26:00 +0800
  • ed006a254f [feat] add DeepSpeed pretrain (ZeRO-2, BF16, auto LR, gradient accumulation) baihaochen 2026-05-22 00:27:06 +0800
  • 6534447b62 [fix] issue#771 jingyaogong 2026-05-19 17:40:03 +0800
  • 4497610ec0 [fix] issue#771 jingyaogong 2026-05-19 17:40:03 +0800