Nov 13 Today

1. 17:40

   tinynext-v2-3b: 12.5M params, 56 min on GPU

   cfahlgren1 HF Staff
   Verified signature

   dc714b2
2. 17:10

   export the compressed browser format from train.py instead of int8 activations

   cfahlgren1 HF Staff
   Verified signature

   4dece7f
3. 17:05

   tinynext-v2-nofim: 12.5M params, 28 min on GPU

   cfahlgren1 HF Staff
   Verified signature

   e5dde63
4. 16:30

   fetch the shared eval from v2/ even when uploading to an experiment folder

   cfahlgren1 HF Staff
   Verified signature

   0de7556
5. 16:26

   v2: 13 MB browser build (int8 download, float compute), export/compress scripts, onnx evals

   cfahlgren1 HF Staff
   Verified signature

   33bfb6d
6. 16:10

   no-FIM models are evaluated with prefix-only prompts

   cfahlgren1 HF Staff
   Verified signature

   793c5c9
7. 15:40

   tinynext-v2: 12.5M params, 28 min on GPU

   cfahlgren1 HF Staff
   Verified signature

   1cffd47
8. 14:43

   prepare data on a CPU job, train from the prepared files

   cfahlgren1 HF Staff
   Verified signature

   63b64fd
9. 14:14

   parallel data prep, data on GPU, trackio, safetensors

   cfahlgren1 HF Staff
   Verified signature

   9bba47b
10. 14:06

    fix step count

    cfahlgren1 HF Staff
    Verified signature

    88610cb
11. 13:47

    v2: training script, eval harness, eval set and baseline metrics

    cfahlgren1 HF Staff
    Verified signature

    3f86878