Independent community conversion of HuggingFaceTB/SmolLM2-135M-Instruct (commit 12fd25f77366fa6b3b4b768ec3050bf629380bac, license apache-2.0). Not an official release of the upstream author.
This is the full-precision BF16 GGUF, not quantized: 0.3 GB in total, as a single file.
| File | Bytes | SHA-256 |
|---|---|---|
Sakura-SmolLM2-135M-Instruct-BF16.gguf |
270886400 | e8dbe545492a548832dbfd8fc178fa61724aaadd9916115da636871a84f1ee46 |
The weights were read from the original safetensors and written to GGUF BF16 by the unmodified official convert_hf_to_gguf.py of llama.cpp (revision 19ff817fb25e9a1b838b1d7f9f4325f7bf9779c4). Every file's SHA-256 was read back from the Hub after the upload.
A byte-identity comparison with a local run and a third-party GGUF was made for one reference model (Qwen2.5-0.5B-Instruct) with another llama.cpp revision (620d2225cc436ab0b001e3200e590e3b286ef9ab); it has not been repeated with this revision or for this model.
No inference test and no quality measurement was done on this file. Smaller quantizations made from it are tested separately.
llama-cli -m <first file>.gguf -p "Hello" -n 32
Needs a llama.cpp build that supports this architecture.
中文说明(点击展开)独立的社区转换版本,并非上游作者的官方发布:HuggingFaceTB/SmolLM2-135M-Instruct(提交 12fd25f77366fa6b3b4b768ec3050bf629380bac,许可证 apache-2.0)。
这是全精度 BF16 GGUF,未做量化:共 0.3 GB,单个文件。权重由官方未修改的 llama.cpp(修订 19ff817fb25e9a1b838b1d7f9f4325f7bf9779c4)的 convert_hf_to_gguf.py 从原始 safetensors 转换而来,上传后已从 Hub 回读每个文件的 SHA-256。未对此文件做推理测试或质量评估。
HuggingFaceTB/SmolLM2-135M-Instruct