{"type":"post","id":"qwen3-5-assembly-qwen3-5-0-8b-cpu-01f599","slug":"qwen3-5-assembly-qwen3-5-0-8b-cpu-01f599","title":"qwen3_5-assembly: Qwen3.5-0.8B CPU inference written entirely in x86-64 assembly, 6.51 tok/s on one EPYC thread","description":"A public GitHub project implements CPU inference for Qwen3.5-0.8B in x86-64 assembly, covering tokenization through decoding and running without Python or an ML framework at runtime.","retrieval_nugget":"A public GitHub project implements CPU inference for Qwen3.5-0.8B in x86-64 assembly, covering tokenization through decoding and running without Python or an ML framework at runtime. Its author reports a 1.50× median decode-speed advantage over one Transformers/PyTorch setup on one CPU thread, but explicitly limits that result to two runs and one prompt.","published_at":"2026-11-08","updated_at":"2026-11-08","record_date":"2026-11-08","date_kind":"published_at","topics":["github"],"entities":["Qwen3.5-0.8B","Qwen3.5 assembly","Hugging Face Transformers","PyTorch","Şuayp Talha Kocabay"],"related_patterns":["harness-architecture-outlives-model-choice"],"novelty_label":"new_fact","what_changed":"A new Qwen3.5-0.8B CPU inference engine is implemented entirely in x86-64 assembly and reports 6.51 tokens per second on one EPYC thread.","prior_refs":["https://newruntime.com/posts/qwenputer-compiles-a-32-bit-risc-v-cpu-822444/","https://newruntime.com/posts/hitting-a-billion-tokens-per-minute-on-one-f36638/","https://newruntime.com/posts/pli-4-0-0-pre-release-open-source-6d6a03/"],"source_url":"https://github.com/suayptalha/qwen3_5-assembly","source_urls":["https://github.com/suayptalha/qwen3_5-assembly"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:qwen3-5-assembly-qwen3-5-0-8b-cpu-01f599","status":"published","visuals":[],"editor":{"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"routes":{"html":"https://newruntime.com/posts/qwen3-5-assembly-qwen3-5-0-8b-cpu-01f599/","markdown":"https://newruntime.com/posts/qwen3-5-assembly-qwen3-5-0-8b-cpu-01f599.md","json":"https://newruntime.com/posts/qwen3-5-assembly-qwen3-5-0-8b-cpu-01f599.json"},"markdown_url":"https://newruntime.com/posts/qwen3-5-assembly-qwen3-5-0-8b-cpu-01f599.md","next_reads":[{"type":"topic","path":"/topics/github/","url":"https://newruntime.com/topics/github/","title":"Github","reason":"Explore the github topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/github.md","json_url":"https://newruntime.com/topics/github.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}