Ophyx 350M v3, GGUF
Ophyx 350M v3 for llama.cpp, a 350M Python code model trained from scratch on free compute.
| File | Size | Largest log-prob gap vs the original |
|---|---|---|
ophyx-350m-v3-f32.gguf |
1.4 GB | 0.010 |
ophyx-350m-v3-f16.gguf |
700 MB | 0.011 |
ophyx-350m-v3-q8_0.gguf |
372 MB | 0.166 |
Converted with llama.cpp b6800. Every file was checked against the transformers model: RoPE base
100,000, identical tokens on seven test snippets, and the same greedy answer, with the largest per-token log-prob
gap over its 66 tokens shown above.
check.json has the details.
Prompt
<|user|>
Write a function `is_palindrome(s)` that ignores case and non-letters.
<|python|>
The answer ends with <|end|>. The tokenizer adds no BOS token, so send the prompt as is. For a
follow-up, put the earlier turns in front, each as <|user|>\n{prompt}\n<|python|>\n{answer}\n<|end|>\n.
Licence
Apache-2.0.