Views
No views yet
docs/ai/dbias_scratch_overrun_silent_zero_bug.md.| Final validation loss | 3.0128 |
| HellaSwag (acc_norm) | 3665/10042 = 36.50% (Wilson 95% CI [35.6%, 37.4%]) |
| Final train loss | 2.9945 |
| Throughput | ~109,535 tok/s (4188.3 ms/step, median over the last 500 steps) |
sample-10BT, GPT-2 BPE, one epoch = 22,345 stepsmodel.safetensors — GPT2LMHeadModel-compatible export, loadable with transformersmodel_22345.bin — the raw llm.mojo/llm.c-format checkpointinfer_gpt2.mojo loads any of a local .bin, a local .safetensors, or this
repo directly via --hf ulmentflam/gpt2-774m-fineweb-nvfp4-mojo.1git clone --recurse-submodules https://github.com/ulmentflam/llm.mojo.git
2cd llm.mojo && make install-cuda
3pixi run python data/fineweb.py -t classic -v 10B -m gpt-2
4make build-bf16 WORLD_SIZE=7