HallD commited on
Commit
ecd8f24
·
verified ·
1 Parent(s): 2954396

Weights size, measured

Browse files
Files changed (1) hide show
  1. README.md +1 -1
README.md CHANGED
@@ -49,7 +49,7 @@ print(response.nouls["billing"].noul, response.choices["tone"].probabilities)
49
  The first call downloads the model and its backbone,
50
  [`Qwen/Qwen3.5-4B-Base`](https://huggingface.co/Qwen/Qwen3.5-4B-Base) at a pinned revision, into the
51
  Hugging Face cache. After that it runs offline (`HF_HUB_OFFLINE=1`). The backbone runs in bfloat16;
52
- a GPU with 12 GB holds it.
53
 
54
  The call shape follows the TypeSafe SDK (0.7.5): `YevClient` and `AsyncYevClient` take its
55
  `system_one(state, questions, model=, assets=, response_model=)`, its question objects (TypeSafe's
 
49
  The first call downloads the model and its backbone,
50
  [`Qwen/Qwen3.5-4B-Base`](https://huggingface.co/Qwen/Qwen3.5-4B-Base) at a pinned revision, into the
51
  Hugging Face cache. After that it runs offline (`HF_HUB_OFFLINE=1`). The backbone runs in bfloat16;
52
+ its weights and the adapter take about 9.6 GB.
53
 
54
  The call shape follows the TypeSafe SDK (0.7.5): `YevClient` and `AsyncYevClient` take its
55
  `system_one(state, questions, model=, assets=, response_model=)`, its question objects (TypeSafe's