Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
C
Merge pull request #1786 from kevin9327/fix/image-url-part-shape
Vincenzo Fornaro6fd47e1ae70c
diff --cc c/openai_server.pyindex 9d356cd5,f5322b27..f8e8302d--- a/c/openai_server.py+++ b/c/openai_server.py@@@ -2080,10 -2062,9 +2095,9 @@@ def expand_qwen38_images(messages, modecontinuekind = part.get("type")if kind == "text":- pieces.append(part.get("text", ""))+ pieces.append(_text_part(part, index, position))elif kind in ("image_url", "input_image"):- url = (part.get("image_url") or {}).get("url") if kind == "image_url" \- else part.get("image_url") or part.get("url")+ url = _image_part_url(part, kind)data = _image_bytes_from_url(url)patches, grid_h, grid_w = _preprocess_qwen38_image(data, model_dir, max_tokens)@@@ -2128,10 -2109,9 +2142,9 @@@ def expand_glm53_images(messages, modelcontinuekind = part.get("type")if kind == "text":- pieces.append(part.get("text", ""))+ pieces.append(_text_part(part, index, position))elif kind in ("image_url", "input_image"):- url = (part.get("image_url") or {}).get("url") if kind == "image_url" \- else part.get("image_url") or part.get("url")+ url = _image_part_url(part, kind)data = _image_bytes_from_url(url)patches, grid_h, grid_w = _preprocess_image(data, model_dir)tokens = (grid_h // 2) * (grid_w // 2)@@@ -2166,10 -2146,9 +2179,9 @@@ def expand_dsv41_images(messages, modelcontinuekind = part.get("type")if kind == "text":- pieces.append(part.get("text", ""))+ pieces.append(_text_part(part, index, position))elif kind in ("image_url", "input_image"):- url = (part.get("image_url") or {}).get("url") if kind == "image_url" \- else part.get("image_url") or part.get("url")+ url = _image_part_url(part, kind)data = _image_bytes_from_url(url)patches, grid_h, grid_w, llm_h, llm_w = _preprocess_dsv41_image(data, model_dir, max_tokens)