Hugging Face's logo

birdhackor
/
tiny-perceptron-course-models

Model card Files
xet
 Community

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Tiny Perceptron：課程教學模型

這些 tiny 模型用來練習從零訓練、下載與推論。各模型的任務、實測結果與限制以其模型卡為準；小型受控任務的結果不能推廣為通用能力或高性能保證。

本頁僅列出課程 docs/course-experiments/public-models.json 中已發布的模型，不代表所有課程實驗都已完成。下載連結固定在各次發布的 HF commit；根 README 的後續更新不改變那些版本。

程式碼採 MIT。權重與資料按檔案適用個別授權；下載前請閱讀各模型的 LICENSE、THIRD_PARTY_NOTICES.md 與 export-manifest.json，不能把程式碼的 MIT 當成所有檔案的權重授權。

已發布模型 模型卡 固定 HF revision 逐檔授權
simple_models 任務與實測限制 fbbff36990db LICENSE / 逐檔清單
text_foundation 任務與實測限制 2ba1278993a6 LICENSE / 逐檔清單
real_text 任務與實測限制 c8416bcf4d54 LICENSE / 逐檔清單
tokenizer 任務與實測限制 58eb946c0eb9 LICENSE / 逐檔清單
sft 任務與實測限制 14293af762e7 LICENSE / 逐檔清單
sft_ablation 任務與實測限制 e30b15712b79 LICENSE / 逐檔清單
style 任務與實測限制 23b58c077a2d LICENSE / 逐檔清單
lora 任務與實測限制 1bfb0ed028d3 LICENSE / 逐檔清單
safety 任務與實測限制 308aa207d691 LICENSE / 逐檔清單
dpo 任務與實測限制 9c3601de9cd4 LICENSE / 逐檔清單
encoders 任務與實測限制 93e45c77a7f6 LICENSE / 逐檔清單
contrastive 任務與實測限制 03cddf7bf0a2 LICENSE / 逐檔清單
projector 任務與實測限制 6d115860fc0a LICENSE / 逐檔清單
vqa 任務與實測限制 e17db92f5504 LICENSE / 逐檔清單
vision_ablation 任務與實測限制 71b6a22c6934 LICENSE / 逐檔清單
ocr 任務與實測限制 ce803a4c1eb7 LICENSE / 逐檔清單
audio 任務與實測限制 1c946fcb7f1d LICENSE / 逐檔清單
joint 任務與實測限制 b7908e156915 LICENSE / 逐檔清單
real_modal 任務與實測限制 0c284d926ee7 LICENSE / 逐檔清單
modern 任務與實測限制 59d993473b51 LICENSE / 逐檔清單
moe 任務與實測限制 17fc5f20686f LICENSE / 逐檔清單
efficiency 任務與實測限制 0e1628d51964 LICENSE / 逐檔清單
precision 任務與實測限制 fb1d740aa140 LICENSE / 逐檔清單
quantization 任務與實測限制 b41d097e4c10 LICENSE / 逐檔清單
qat 任務與實測限制 30ea5592953d LICENSE / 逐檔清單
distillation 任務與實測限制 fa69713c7dec LICENSE / 逐檔清單
multimodal_distillation 任務與實測限制 ae3c4263c4f9 LICENSE / 逐檔清單
rag 任務與實測限制 660fa1d1fa67 LICENSE / 逐檔清單
tools 任務與實測限制 973d02736f4e LICENSE / 逐檔清單
reasoning 任務與實測限制 ac5ac599faab LICENSE / 逐檔清單

在課程 checkout 下載與推論

使用本課程程式碼與配對的推論腳本。下載器只取選定模型，匿名下載並核對固定 revision、檔案大小與 SHA-256。

uv sync --extra cpu
uv run --extra cpu python scripts/fetch_course_models.py --list

下面列出公開清單中已提供的推論設定；生成輸出仍需核對，不從下載成功推論模型能力。 RAG 的文字推論範例直接提供文件；工具模型的文字推論只產生請求，沒有自行執行計算。檢索、嚴格驗證、工具執行與結果回填仍是外層程式的工作；完整操作請依各模型卡的流程。 OCR 與真實媒體組下方的合成形狀／單音輸入只檢查程式入口，不能用來評估數字辨識或人聲辨識。這兩組請依模型卡準備數字圖片或固定媒體，再使用其 --image／--audio 範例。

simple_models 

uv run --extra cpu python scripts/fetch_course_models.py --model simple_models
uv run --extra cpu python scripts/infer_simple.py checkpoints/course/simple_models/mlp3.pt --prompt '顏色=' --tokens 24 --device cpu

固定版本權重：

  • bigram.pt · 3,618 bytes · SHA-256 2a51d84afd66eb2a3d528d2447cfe978e8bbbadd2faf451525f50b900d8bddd6
  • mlp1.pt · 6,584 bytes · SHA-256 8eecee287fbd149e5d1202d16f9e6cbcf6de20f93ac38db2f4cb1075a96836a0
  • mlp3.pt · 8,632 bytes · SHA-256 9de6d06eb4ef9894562e7565d917db913b590cd7663f4510c74c17f4cf5b7332
  • mlp5.pt · 10,680 bytes · SHA-256 b4563451e72b61fdb0cdc91809908ae8157751dd511c66d4a4cc53559d4c772b

text_foundation 

uv run --extra cpu python scripts/fetch_course_models.py --model text_foundation
uv run --extra cpu python scripts/infer.py checkpoints/course/text_foundation/model.pt --prompt 'color=blue;shape=circle;' --tokens 32

固定版本權重：

  • model.pt · 575,494 bytes · SHA-256 98bfedd6cf596a3bb90a21f4acd67fd9c4c2caf67435b2a381b6cfc11ed80834
  • scaling-w16-n16.pt · 62,376 bytes · SHA-256 e449a2480196ab0cc80a324ec894d7fae24cdf02204169e33c08ffe7ab067889
  • scaling-w16-n64.pt · 62,376 bytes · SHA-256 67d867a814f4254e28db07ec8c53ef62ea2eb842d992e79a9d01a510dd17ab63
  • scaling-w32-n16.pt · 141,928 bytes · SHA-256 7a1d7c93a7cc275d2eb789d893103e4cd420058559902ba4feba877aa0cabfad
  • scaling-w32-n64.pt · 141,928 bytes · SHA-256 5cb931b7857c4204b7634334b284da7322dd230fec48aca93ebdbe107e8f5af4
  • start.pt · 575,494 bytes · SHA-256 90a6a6dd093aafd3cf8588c7db76647aaf50f48fe643d117a9d161e301b191a0

real_text 

uv run --extra cpu python scripts/fetch_course_models.py --model real_text
uv run --extra cpu python scripts/infer.py checkpoints/course/real_text/tinystories.pt --prompt 'Once upon a time' --tokens 32 --temperature 0.0 --device cpu

固定版本權重：

  • chinese-poetry.pt · 577,729 bytes · SHA-256 66090e65760ee09450c920563c837c568534684902f90cf98d1f6aeacc2b96ae
  • tinystories.pt · 577,560 bytes · SHA-256 ca6d734aa3fe2eb1f97196982fd057ba6bd14078579d43d4756991621a54b7ed

tokenizer 

uv run --extra cpu python scripts/fetch_course_models.py --model tokenizer
uv run --extra cpu python scripts/infer.py checkpoints/course/tokenizer/bpe512.pt --prompt Once --tokens 32 --temperature 0.0 --device cpu --tokenizer checkpoints/course/tokenizer/tokenizer-bpe512.json

固定版本權重：

  • bpe512.pt · 237,081 bytes · SHA-256 5e1d520fdd97ddc0e1ae15de06e5e596dd19d0f5d5e6d4edc00a0f0e6857d3cc
  • byte256.pt · 174,128 bytes · SHA-256 40d7b2e53c55e8b7c36b44e137c5ef40e3e62b75afd6193fb384e8d546a36ee2

sft 

uv run --extra cpu python scripts/fetch_course_models.py --model sft
uv run --extra cpu python scripts/infer.py checkpoints/course/sft/model.pt --prompt 'color=blue;shape=circle;pitch=low;shape?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • model.pt · 575,622 bytes · SHA-256 751fe9448d209fc34fcd723ee69aa1f10fe40d8c38c2c689c11decfae8f4e132
  • pretrain-sft.pt · 577,723 bytes · SHA-256 5a157b8b0c767f3b7041e6bd1f2a47826649b270412a7b9ecce5d9443326d6d3
  • pretrain.pt · 577,583 bytes · SHA-256 bb68217d83d197e59dd6cf3a8a2cb4eb73a1dba703c38d9805ed0abc6be62817
  • ultrachat-pilot.pt · 158,440 bytes · SHA-256 d2e7e6b9fab8e6d111d446f5253cc85a71e42b84360c42e7dd592373fb0ae888

sft_ablation 

uv run --extra cpu python scripts/fetch_course_models.py --model sft_ablation
uv run --extra cpu python scripts/infer.py checkpoints/course/sft_ablation/replay.pt --prompt '1+2=?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • b-only.pt · 575,593 bytes · SHA-256 bd1c2ff521093a50c571dbbceffd876be029803ce0aba7cac92358a9e92adcf5
  • clean.pt · 575,558 bytes · SHA-256 d947d654d84c5bca5324f2ea5fefa8072ecedb67da77b93f75f8a3c6a9471ea8
  • noisy.pt · 575,558 bytes · SHA-256 e00105d154e84078aa69770f925ba55c7930cdb00bfc22fbb96cced5349de28f
  • replay.pt · 575,593 bytes · SHA-256 e94b7f364917b48249081f9a87f3418b49cfa8cf644bed83af18754c7ecbb39e

style 

uv run --extra cpu python scripts/fetch_course_models.py --model style
uv run --extra cpu python scripts/infer.py checkpoints/course/style/model.pt --prompt 'style=json; 1+2=?' --tokens 96 --chat --temperature 0.0 --device cpu

固定版本權重：

  • content.pt · 576,908 bytes · SHA-256 57a49e9bcdfb1e411d9a64c683bae90ad1fba41072bace0f1bd68e0e1f3a35f6
  • default-concise.pt · 577,764 bytes · SHA-256 5ff737a4376c42b675f9198d9f1e0cfb888c412cdf54d203cbd474861c5924a7
  • default-vivid.pt · 577,694 bytes · SHA-256 7b48a2e8e1e70fb016f00886cdf949d81cc9b6952d530c2a7fbc33851b96e875
  • model.pt · 575,558 bytes · SHA-256 a3481cd6ee5415563f107d6ec07aeecba92f2520ab7d45fdc5373840f8a78607

lora 

uv run --extra cpu python scripts/fetch_course_models.py --model lora
uv run --extra cpu python scripts/infer.py checkpoints/course/lora/merged-vivid.pt --prompt '1+2=?' --tokens 96 --chat --temperature 0.0 --device cpu

固定版本權重：

  • adapter-concise.pt · 46,721 bytes · SHA-256 a17192b40050ea891063a4e29255fee3c77783800be58b5187662853337703fa
  • adapter-vivid.pt · 46,665 bytes · SHA-256 dd2525010901218db8bd90fa5ab0e863e10424b2bddb8ec13515bc4c49abbfbe
  • full-sft.pt · 577,519 bytes · SHA-256 b9fdce59a0ba96e270709110aedc97b00ee98275db8631e193d2caf7ba3c9406
  • merged-concise.pt · 577,921 bytes · SHA-256 1e6059dd4a56c5729f906670b24cbc35012267ca9bb79f78d91080d8540551ad
  • merged-vivid.pt · 577,787 bytes · SHA-256 30c82b7a8d50c8429b623394eccbc2f36940c1b15403167308e42abbe23cfbe6

safety 

uv run --extra cpu python scripts/fetch_course_models.py --model safety
uv run --extra cpu python scripts/infer.py checkpoints/course/safety/model.pt --prompt '盒子0；permission=True；請提供秘密碼。' --tokens 128 --chat --temperature 0.0 --device cpu

固定版本權重：

  • model.pt · 575,558 bytes · SHA-256 0271ddaf1320059110cfaf0e10d0477613f510b741f4299b7976250b36cf09a8
  • safety-only.pt · 577,624 bytes · SHA-256 719316bb22c664d260321a1a151ec6c9f00cfb5b47689a6897c369c9cea69e5e

dpo 

uv run --extra cpu python scripts/fetch_course_models.py --model dpo
uv run --extra cpu python scripts/infer.py checkpoints/course/dpo/model.pt --prompt '1+2=?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • beta1.pt · 575,558 bytes · SHA-256 e25fea870fa479d7acdbad29cdb91d716cf741c754470cbec72793023912d1a2
  • format-model.pt · 577,659 bytes · SHA-256 a8b10e2de373d50c94e209aa12302c04c29cd25071594009c0de110da0474c2d
  • model.pt · 575,558 bytes · SHA-256 278c537de53c2f991a040c9edd7bc103a91cfd25a501e73a0b6b777881ec7a73
  • ultrafeedback-pilot.pt · 158,468 bytes · SHA-256 e7a8927fc5d96232f8514f05bca55b5656273d86a63bd34bdd1383eb7baa801a
  • ultrafeedback-sft.pt · 158,422 bytes · SHA-256 242e5753ebf11027e7c1fa2f5cf9dd08df9c055490d1ed8ecb87c33fc8cdf563

encoders 

uv run --extra cpu python scripts/fetch_course_models.py --model encoders

此項公開清單未指定推論 CLI；請依模型卡的使用設定與 architecture 載入。

固定版本權重：

  • audio.pt · 6,385 bytes · SHA-256 23e4c805224191e9745c35b338bd745944984c3cae7621930179db7a66025a79
  • vision.pt · 10,106 bytes · SHA-256 a94424e267950bd2e253aa00231613f7bbc19eb10f2f6a4a9154ca05a98333ec

contrastive 

uv run --extra cpu python scripts/fetch_course_models.py --model contrastive

此項公開清單未指定推論 CLI；請依模型卡的使用設定與 architecture 載入。

固定版本權重：

  • one_way.pt · 26,253 bytes · SHA-256 6ab3643d6a0f041982db2cf3103896d198644f1c11429adbb09e052b854c7691
  • two_way.pt · 26,253 bytes · SHA-256 77a8d58118c9395756ba691266df8d57e160398969ac6261d367eb3e9b28c5c8

projector 

uv run --extra cpu python scripts/fetch_course_models.py --model projector
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/projector/model.pt --device cpu --tokens 16 --prompt describe --color red --shape square

固定版本權重：

  • model.pt · 597,105 bytes · SHA-256 09e7bedda6145f190b5df41184ada2834517fa6d8ec01b8593a7615f2039b5ea

vqa 

uv run --extra cpu python scripts/fetch_course_models.py --model vqa
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/vqa/all.pt --device cpu --tokens 16 --prompt 'shape?' --color red --shape square

固定版本權重：

  • all.pt · 597,001 bytes · SHA-256 c623e0195d41224ec44bf99fdb56a070b4dfd3235e44682c8dabba7a41f36b2f
  • all_replay.pt · 600,309 bytes · SHA-256 6a5634a802798201ea2f9b25dc6910980b9e69e50f33a05e845d0462c87be7de
  • direct_vqa.pt · 600,309 bytes · SHA-256 bf703a40a3da1d6b508ca3bda31e987cbf4482e85dcdc08899c3d4a8011f1e5e
  • partial.pt · 599,577 bytes · SHA-256 e8f998aa89c3d94caa5a6ceb9270e4dcd3872e17517fdd548711669533895bd4
  • projector_only.pt · 600,581 bytes · SHA-256 c9ea03a0810a3d17aa16e3d98083f46a20ec8dc996570e7822bf3d20ebca832c

vision_ablation 

uv run --extra cpu python scripts/fetch_course_models.py --model vision_ablation
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/vision_ablation/model.pt --device cpu --tokens 16 --prompt 'shape?' --color red --shape square

固定版本權重：

  • model.pt · 597,105 bytes · SHA-256 e206c6e5956c75de6505899eadde291706969eb76d548b025ca4be36330cee2d
  • patch-4.pt · 599,577 bytes · SHA-256 f867ca9aa401ae3efe69aa764dc8f214f9587ac392861d295c30950468b91a91
  • patch-8.pt · 608,025 bytes · SHA-256 4eadde6b99d7345d404756a420a042ae0fee3ddc51066b7e5c39ab54083aec3f

ocr 

uv run --extra cpu python scripts/fetch_course_models.py --model ocr
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/ocr/model.pt --device cpu --tokens 5 --prompt 'read digits' --color red --shape square

固定版本權重：

  • model.pt · 597,105 bytes · SHA-256 ffbec1ca9affb7c53ea3133464f3b60a84c315cf2e4f8dadb108f8b3876a0dcf

audio 

uv run --extra cpu python scripts/fetch_course_models.py --model audio
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/audio/model.pt --device cpu --tokens 6 --prompt 'pitch?' --frequency 220.0

固定版本權重：

  • model.pt · 597,105 bytes · SHA-256 76c271934541c7005d3b1f450697a1d673cb827bab22fe015fb074a57df8d950

joint 

uv run --extra cpu python scripts/fetch_course_models.py --model joint
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/joint/model.pt --device cpu --tokens 16 --prompt 'joint?' --color red --shape square --frequency 220.0

固定版本權重：

  • model.pt · 597,105 bytes · SHA-256 2d8c41239c0fd08ae8fdf3128dd05effa54ac7f76960bed4fc963eb956b2339a

real_modal 

uv run --extra cpu python scripts/fetch_course_models.py --model real_modal
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/real_modal/fsdd.pt --device cpu --tokens 32 --prompt 'digit?' --frequency 220.0

固定版本權重：

  • fashion-mnist.pt · 600,529 bytes · SHA-256 2c9a795b838fb06bb287ce704570588783d42470a71e51ddf444d3a34da545a8
  • fsdd.pt · 609,341 bytes · SHA-256 a80abb51f06dc935e35f5fcede478474bf4ede11b9e250e16421abedecbb3fb5

modern 

uv run --extra cpu python scripts/fetch_course_models.py --model modern
uv run --extra cpu python scripts/infer.py checkpoints/course/modern/baseline.pt --prompt 'Once upon a time' --tokens 32 --temperature 0.0 --device cpu

固定版本權重：

  • baseline.pt · 577,583 bytes · SHA-256 29790d1c2ded1863d2dcda66665a8176648055078fb185611380633513fce0d7
  • relu2.pt · 575,622 bytes · SHA-256 e08dbabca1880f3e1112a06526a69413f9e7dd450a3422abc165e6f8e974d9f4
  • rmsnorm.pt · 574,171 bytes · SHA-256 6cf49f2b23dd7b5887e4a630a6b96126bbc65402a07db0d8fdcf0fda0266e273
  • rope.pt · 542,569 bytes · SHA-256 d92168652cf55fa9b7335af91e5d2bc84e70e0684b0c9797407bdbbe2335d568
  • swiglu.pt · 709,785 bytes · SHA-256 a4bcd1f974a64c7b87264478308344f56e8ef295e1be236272a62b31407a438d
  • tied.pt · 507,881 bytes · SHA-256 9648792bc322eba668baf7b93c1800f2093451dc8f5c5a698711f6454a50edf5

moe 

uv run --extra cpu python scripts/fetch_course_models.py --model moe
uv run --extra cpu python scripts/infer.py checkpoints/course/moe/top2_aux0.01.pt --prompt 'Once upon a time' --tokens 32 --temperature 0.0 --device cpu

固定版本權重：

  • dense_active_top1.pt · 577,898 bytes · SHA-256 7740bb89da376e8b8c335086c72bfe128cada4548bbf00ff77cfa2611c36c292
  • dense_active_top2.pt · 842,346 bytes · SHA-256 e20da254f9aed4a803e46c8b368c178aa08ec881fb04c40a595c6993ab93abcb
  • dense_total.pt · 1,330,584 bytes · SHA-256 24bd49a51ccf7d4e3d34b91709b13f02f7d178c17e7e9fe9d2b76d56775af2e0
  • top1_aux0.01.pt · 1,383,215 bytes · SHA-256 8a97133cbf9ab9f5d2f75864d98dc86f66214560b2a155548065ada7b44c727b
  • top1_aux0.pt · 1,383,032 bytes · SHA-256 f3df1cd360f2341f24def64e19d84241362ab14e518201d693b6da08521c5e21
  • top2_aux0.01.pt · 1,383,215 bytes · SHA-256 6ecad573c367f46e6eab4a5a7a08ac7917706d1ee159931d4e8e950163f2f3c7
  • top2_aux0.pt · 1,383,032 bytes · SHA-256 42009e9265cf89b5de7de208ec8468960599e4da9005c19af4938b4488cb56d2

efficiency 

uv run --extra cpu python scripts/fetch_course_models.py --model efficiency
uv run --extra cpu python scripts/infer.py checkpoints/course/efficiency/ordinary.pt --prompt 'color=blue;shape=circle;pitch=low;shape?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • accumulated.pt · 577,688 bytes · SHA-256 91ec3c9467448201e5df31d7241257c6966cd9dee9483f1736633794c7474eb5
  • activation_checkpoint.pt · 578,038 bytes · SHA-256 e1fafe2efa762585d18f88af06b7318a34bfd2d0e8c431d11425b02bd2fc6517
  • gqa.pt · 526,400 bytes · SHA-256 f37fc52d5bea892acee79c73d794798396bfa20d2070f5386b05aad21a213a80
  • mha.pt · 575,552 bytes · SHA-256 db869d795edc0fa24fcc04025db29685f151ff0bd5b9a963981e2907d529064b
  • ordinary.pt · 577,583 bytes · SHA-256 28e6eb93acc4ced4efb19fc77f620dcc032a08ba753fe64d3cc6e6465245237c
  • packed.pt · 575,657 bytes · SHA-256 630ef043d062a62202bbdafc7098cdc53ad434de128f0c051b96d79e38f0022d
  • padded.pt · 575,657 bytes · SHA-256 bb53820720a43ee534e2fc91fee16645dbc814c02b4607693c0132a94d1c6e0a
  • sdpa.pt · 575,587 bytes · SHA-256 65078eb0fd98d4cdcf29bc89cb308cf17715a855c3fcfc705755172d70f1fc22

precision 

uv run --extra cpu python scripts/fetch_course_models.py --model precision
uv run --extra cpu python scripts/infer.py checkpoints/course/precision/fp32.pt --prompt 'color=blue;shape=circle;pitch=low;shape?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • bf16.pt · 575,523 bytes · SHA-256 741c0d63882442d547e74bc484d67ce78364a60212a4f567202037472ce4ffad
  • fp16.pt · 575,523 bytes · SHA-256 20daeb4fbde13a7374ea823346664b37075b5d19a14427615c732c148a09bcce
  • fp32.pt · 575,523 bytes · SHA-256 d3c6edc501b94d0288387b8792f9d9f0098b4033524495839917743454f42c19

quantization 

uv run --extra cpu python scripts/fetch_course_models.py --model quantization
uv run --extra cpu python scripts/infer.py checkpoints/course/quantization/model.pt --prompt 'color=blue;shape=circle;pitch=low;shape?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • fp32.pt · 575,587 bytes · SHA-256 74e8bab39917102d92fd988470dde4281da2a3e647a115b4e05e5edee6f058c1
  • model.pt · 181,381 bytes · SHA-256 17e7f0e824b8bebb8396c9384a1887463b30575f337ac88e1072f30177da41cb
  • packed8.pt · 241,189 bytes · SHA-256 27b487ce26e4da4ec7f406269a59d5cbf551064b94d32ab95ef1b10eb25bf0a0

qat 

uv run --extra cpu python scripts/fetch_course_models.py --model qat
uv run --extra cpu python scripts/infer.py checkpoints/course/qat/model.pt --prompt 'color=blue;shape=circle;pitch=low;shape?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • fp_finetuned.pt · 577,723 bytes · SHA-256 becc1eb7247cfb867140f56cb8ad711e247127967f5ca745c1ce170a0a1ac708
  • initial_fp32.pt · 577,787 bytes · SHA-256 3f7103f8b8df0cfe0712866964a1e9a6116b271000f6803068c7fc2043e086bc
  • initial_ptq4.pt · 184,341 bytes · SHA-256 17d9db0bbd66eced55800a2b338a0a2ad8967844feb3477824b4218affa26a27
  • matched_ptq4.pt · 184,341 bytes · SHA-256 0de4b8d25cf9f3e5caa2b5d342523ac3928cdf449262ff3ade7f4c3f27b34008
  • model.pt · 181,381 bytes · SHA-256 d4753c4d944bf2629f8e589a76f2bc809c996c0002628238f9f9153a0dcdeedf
  • qat_float.pt · 577,618 bytes · SHA-256 af2ab595d987dc404fa30795d518843e824a2533e90841ae11b379f187e056a0

distillation 

uv run --extra cpu python scripts/fetch_course_models.py --model distillation
uv run --extra cpu python scripts/infer.py checkpoints/course/distillation/sft-w32-ce_kl.pt --prompt 'color=blue;shape=circle;pitch=low;shape?' --tokens 32 --chat --temperature 0.0 --device cpu

固定版本權重：

  • moe-teacher.pt · 1,383,218 bytes · SHA-256 a4efa3a172910542052b10c8b71412e090d22625fe0d1d8ee19531ac69812b5a
  • moe-w32-ce.pt · 141,941 bytes · SHA-256 a58d0a915e15574bbb1f88def20c5931f54a01564be4fc8ee18bf9a0e3d6ea5a
  • moe-w32-ce_kl-packed4.pt · 74,111 bytes · SHA-256 91155d5e00478420bd478cddd0800d210def584c01c43ce09aeb7e45c654f2b4
  • moe-w32-ce_kl.pt · 142,010 bytes · SHA-256 11d2251b976ed91bdbe9c0da415eeabf8d34bb4a940e722b3ff9997364d9bb2e
  • sft-teacher.pt · 577,752 bytes · SHA-256 0a0d86ee5ca383a7c1fdf185eaf2c23d6af90fe3c6fa22dec437967ac246f028
  • sft-w16-ce.pt · 62,389 bytes · SHA-256 f2e8b39bc982c4847e3ab022ffd83bedfac5237a63aa53f074a6747cf8aafe21
  • sft-w16-ce_kl-packed4.pt · 41,023 bytes · SHA-256 998573ab716e6c873875c06bbf296738d91136fdcc81be184a763dfa5a234a90
  • sft-w16-ce_kl.pt · 62,458 bytes · SHA-256 2013439c43a02f1d970ed538f5001e590afbc61c6163bde37608c64712e785e2
  • sft-w16-teacher_hard.pt · 62,619 bytes · SHA-256 63118322858e2cf1ec563e8c064059c2210dfd8f976b3a8ee0cc7df6ba578eb4
  • sft-w32-ce.pt · 141,941 bytes · SHA-256 a727a99bbc0bee44744ba0deb387490b8921fb6def8f88184a4d39113b25e888
  • sft-w32-ce_kl-packed4.pt · 74,111 bytes · SHA-256 e5a2bde7122991cfab7c7f28f59793695e8f5ce5cd12bb662d2ecea527bde65b
  • sft-w32-ce_kl.pt · 142,010 bytes · SHA-256 addc52576558ccead03851d86e748873bdb52c9b13fd7b30f34a082c2d453f0e
  • sft-w32-teacher_hard.pt · 142,171 bytes · SHA-256 cbca0d92444150baf75820dba01d5e64e5a11d5459b8387466fdcce78b9c2fcb
  • style-teacher.pt · 577,822 bytes · SHA-256 1ac6baf3ea200aa407946e29d372b677eacdc510e8490cfd52e52fe3a1e7bfa0
  • style-w32-ce.pt · 141,987 bytes · SHA-256 ba5dcf0f6e06a0be495366a7e959fb672e5a60c2679f377eb58b7ad7e9f0e270
  • style-w32-ce_kl-packed4.pt · 74,235 bytes · SHA-256 24345ebe6b5b1e3a227a1edab2fa8038dc1a3f3830abf3aa5b2bde61664db3a9
  • style-w32-ce_kl.pt · 142,056 bytes · SHA-256 ca85ea2a6dcfe2ae20fb045122263af581c0f26d0b4e260723a43d6807404f48
  • style-w32-teacher_hard.pt · 142,281 bytes · SHA-256 f07f6ad0c139d98c84376391f16417c63e5137fcdd033b63a70a631a49f77736

multimodal_distillation 

uv run --extra cpu python scripts/fetch_course_models.py --model multimodal_distillation
uv run --extra cpu python scripts/infer_modal.py checkpoints/course/multimodal_distillation/joint-ce_kl.pt --device cpu --tokens 16 --prompt 'joint?' --color red --shape square --frequency 220.0

固定版本權重：

  • joint-ce.pt · 157,605 bytes · SHA-256 3969e8912358d070384f56c436a47d19fcf21e0a11322c4b79755de738a4dc41
  • joint-ce_kl.pt · 157,725 bytes · SHA-256 b8a04ef072275087e3e01b4201986f409c83d97afd2033b5bcbc4da74ed96761
  • joint-teacher.pt · 600,657 bytes · SHA-256 a0165c8108fb895e0987c87465ff0f4cb845693c039f411567dfccdfd816640f
  • vqa-ce.pt · 155,861 bytes · SHA-256 5e3e84d83704e469a22ffdbd212f61eca7c3efb8f81e08c3b13c342356f73e63
  • vqa-ce_kl.pt · 157,645 bytes · SHA-256 5b60922fe03249a9f643c4ba89140f9108841ec9a52084a8fbefb58471f8e153
  • vqa-teacher.pt · 600,553 bytes · SHA-256 7f4f3797ab18cd2e7b0790f7bf3cd4eddd4f9039be6de8c7e8771681264474bd

rag 

uv run --extra cpu python scripts/fetch_course_models.py --model rag
uv run --extra cpu python scripts/infer.py checkpoints/course/rag/model.pt --prompt 'Docs:
[D0] K000 address=A1
Find:K000
Reply:address[source] or UNKNOWN' --tokens 64 --chat --temperature 0.0 --device cpu

固定版本權重：

  • model.pt · 583,814 bytes · SHA-256 37f4a570e878a85949d7670d5da38148779811dcc4cc5bb3f2df0e5fb7e5b194

tools 

uv run --extra cpu python scripts/fetch_course_models.py --model tools
uv run --extra cpu python scripts/infer.py checkpoints/course/tools/model.pt --prompt 'CALC:add(2,3)' --tokens 64 --chat --temperature 0.0 --device cpu

固定版本權重：

  • model.pt · 641,158 bytes · SHA-256 c8789df280e4a4cba06994ee94f1e96c7199b30b9806dc005dd9c32d32d4bdbf

reasoning 

uv run --extra cpu python scripts/fetch_course_models.py --model reasoning
uv run --extra cpu python scripts/infer.py checkpoints/course/reasoning/steps.pt --prompt '(1+2)+3=?' --tokens 48 --chat --temperature 0.0 --device cpu

固定版本權重：

  • direct.pt · 575,657 bytes · SHA-256 2724ea905a92871e2d8216a57408191de6437b4bd78fb37f6b134615c371101a
  • gsm8k-pilot.pt · 141,964 bytes · SHA-256 9070fabfa6cc62e2e1cd923fc148a6eb7c11556b5b8c9fccf93bfcd92e4f8ada
  • policy-strict.pt · 10,521 bytes · SHA-256 01eaa3895041a8c17001a700089b6a476ab1af8aed19f66e9763f8dae09cbd94
  • policy-weak_proxy.pt · 10,561 bytes · SHA-256 3247f10643a85f9c121ecad47240f80833317d859efa34d207cae6b89842d650
  • steps.pt · 575,622 bytes · SHA-256 55f2c3b54a76c4a2b65feb24c8e0ee4b36ba3415f00999fba201e75ec81d91c3
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support