Skip to content

Commit a034a27

Browse files
committed
data: Llama 4 Scout BF16-direct shards (full downloads, 78.4 MB total)
Replaces partial old-pipeline shards with complete BF16-direct + F64x8 runs. All 5 shards streamed from HuggingFace with segment cache + retry. Shard 1: 22 MB (layers 0-10 + embeddings, 117 tensors) Shard 2: 12 MB (layers 11-21, ~110 tensors) Shard 3: 24 MB (layers 22-32, 126 tensors) Shard 4: 13 MB (layers 33-43) Shard 5: 7.4 MB (layers 44-47 + output, 40 tensors) OpenChat: 41 MB (7B Q8_0, 226 tensors) Peak RAM: 134 MB. Compression: Attention 361×, FFN 5939×, Embedding 23770×. https://claude.ai/code/session_01Y69Vnw751w75iVSBRws7o7
1 parent a1d5049 commit a034a27

5 files changed

File tree

10.3 MB
Binary file not shown.
4.05 MB
Binary file not shown.
17.8 MB
Binary file not shown.
8.27 MB
Binary file not shown.
0 Bytes
Binary file not shown.

0 commit comments

Comments
 (0)