Tinker-Stack/DeepSeek-V4-Flash-0731-REAP-150b-TQ3_4S-GGUF

Add experimental disclaimer banner + tag

Move benchmarks to top with base 284B + REAP FP8 reference scores; correct comparability caveat (FP8 ref was thinking-off)

Full benchmark methodology disclosure: per-row protocol, both-protocols matrix, spiral rates, comparability caveat

MBPP+ thinking result: 0.807 (no-think 0.8624 retained); full protocol disclosure per task

Update README.md

Update README.md

MATH-500 thinking: 0.9618 (+24.78 over FP8); protocol note on thinking-mode comparisons

HumanEval+ thinking mode: 0.9207, above FP8 reference (+2.44)

Update model card: capability benchmarks vs FP8, fix stale file refs, add reasoning-mode serving pitfalls

Rename DeepSeek-V4-Flash-0731-REAP-150b-TQ3_4S-GGUF.txt to weights.txt

Rename apex-dsv4-v3b.txt to DeepSeek-V4-Flash-0731-REAP-150b-TQ3_4S-GGUF.txt

Rename apex-v4-tq34s-v3b.gguf to DeepSeek-V4-Flash-0731-REAP-150b-TQ3_4S.gguf

Update README.md

Upload README.md with huggingface_hub

Delete files ds4-flash-heretic-f4-t256-lora-r2.gguf with huggingface_hub

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

Upload folder using huggingface_hub

Update README.md

Upload README.md with huggingface_hub

initial commit