PaoAI/GLM-5.3-Flash-PaoAI-ROCmFP4-STRIX-BALANCED-GGUF
Card update: re-measured on latest version (strix-main 4a8440f58) - decode faster at every depth, chain re-run stated honestly, requirements add paoai-strix-engine route
Sweep section: dual-arm table, depth-4-vs-6 findings, acceptance caveat
Sweep chart v2: dual-arm (draft depth 4 vs 6) + acceptance panel
Add 128K deep-context sweep: results table, chart, plain-read findings
Add deep-context sweep chart (128K: decode, prefill, MTP acceptance)
add fleet recipes repo link
fix: build source points to guevae2/ROCmFPX glm5next branch (3345156) - was unreachable kingjones main ref
Add plain-words serving features table (MTP/FA/KV-q8/reasoning-budget explained)
Add Requirements section: which llama.cpp fork/commit builds this model (glm5next + ROCmFP4 type 101)
Update model card: chain-test N=3 median results (replaces old battery scores)
Upload GLM-5.3-Flash-PaoAI-ROCmFP4-STRIX-BALANCED.gguf with huggingface_hub
Upload GLM-5.3-Flash-PaoAI-ROCmFP4-STRIX-BALANCED.gguf with huggingface_hub
Upload LICENSE with huggingface_hub
Upload README.md with huggingface_hub
initial commit