v75 (S24/SM8650, soc 57): Qwen3-0.6B W8-decode + fp16-prefill bundle — Plane A 6/6
#2
by flying101 - opened
W8 decode + fp16 prefill + W8 lmhead + fp16 embed. Plane A: PASS 6/6 on S24.
flying101 changed pull request status to open
Adds the Hexagon v75 bundle under .
- Files: qwen3-0.6b-1024final.json, qwen06_decn1k_w8.bin, qwen06_pf_f16.bin (prefill), qwen06_lmh_w8.bin, qwen06_embed_f16.bin, tokenizer.json
- Precision: W8 decode + fp16 prefill + W8 lm-head + fp16 embed
- Compile: QAIRT 2.47.0, v75/soc 57; export gate 8/8 match HF greedy
- Device (Plane A, S24): 6/6 standardized prompts, thinking-mode output parsed correctly
flying101 changed pull request title from Add Hexagon v75 (Galaxy S24 / SM8650, soc 57) bundle to v75 (S24/SM8650, soc 57): Qwen3-0.6B W8-decode + fp16-prefill bundle — Plane A 6/6
v75 bundle details: W8 decode + fp16 prefill + W8 lm-head + fp16 embed. QAIRT 2.47, v75/soc 57, export 8/8. Plane A on S24: 6/6 prompts, thinking-mode parsed correctly.
sanmonga22 changed pull request status to merged