<h1 id="base-llm-unmodified-from-official">Base LLM Unmodified from Official</h1><p><em>Since Comfy has not posted Convrot versions of the base trainings. I am posting them here.</em></p><p></p><ul><li><p>These are from the official huggingface shards.</p></li><li><p>Unmodified Conversions to INT8 Convrot.</p></li></ul><p></p><h3 id="llm-guide">LLM Guide</h3><p></p><ul><li><p>QWEN_3_8B - Flux Klein9B</p></li><li><p>QWEN_3_4B - Flux Klein4B, Z-Image</p></li><li><p>QWEN3_06B - Anima</p></li><li><p>Qwen3vl_4b - Krea2</p></li><li><p>T5-XXL - Flux.1 Dev, Schnell, Kontext, Fill, Krea</p></li><li><p>UMT5-xxl - Wan2.1, Wan2.2, T2V, I2V, 3B, 5B and 14B</p></li></ul><p></p><h3 id="converted-by-comfy">Converted by Comfy</h3><ul><li><p><a rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/tree/main/text_encoders">QWEN3vl_32B - MiniMax-H3</a></p></li></ul><h3 id="unsupported-by-comfy">Unsupported by Comfy</h3><ul><li><p>Ministral-3-3B - Ernie Image (CTQ can quantize but comfy rejects the format)</p></li></ul><p></p>

Comfy INT8 Unmodified LLM (Anima, Krea2, Flux, Wan, Z-Image)

qwen3vl_4b_int8_convrot.safetensors

Qwen3vl_4b

Files with SHA256: 3cb5ac15cd2fb2e0948fcb61add384d7192b470b8b88a4d97f0efd1766c169df