IsValorum/Qwen3.8-35B-A3B-Distill-MTP-APEX-I-MiniPlus-V2.1-Abliterated-GGUF
Show direct Ko-fi URL
Number navigation and refine optional support
Complete navigation index and remove duplicate support block
Move navigation to top and clarify optional support
Add optional Ko-fi support link
Correct fidelity tiers against measured delta perplexity
Add compact model-specific V2.1 comparison table
Document Q8_0 and Q4_0 MTP draft heads
Add dedicated Q4_0 MTP draft head for low-VRAM speculative decoding
Add Qwen3.6 Abliterated V2.1 to lineup links
Identify MiniPlus V2.1 edition in lineup links
Clarify established 35B MoE lineup and cross-links
Expand V2.1 model cross-promotion
Promote established V2.1 APEX-I-MiniPlus releases
Add system RAM inference guidance
Restore independent benchmark references; neutralize quantizer attribution
Remove third-party user references from model card
Clarify EmperoAI acknowledgement and Distill V2 status
Add upstream long-generation limitation notice
Keep WikiText-2 results while removing evaluation parameters
Remove exact RAM-offload and laptop throughput figures
Fix nested emphasis in Q5-Q6 fidelity statement
Emphasize Q5-Q6 fidelity at lower-than-Q3_K_M weight
Update model-specific Q3_K_M size reference (17.7 GB)
Tie Q5-Q6 fidelity callout to model result
Make quality comparison model-specific
Clarify external benchmark attribution
Update subtitle to reflect uncensored directional refusal ablation and companions
Fix callout formatting: eliminate broken blockquote splits and align solid alert cards
Validate and format official Hugging Face ModelCard YAML metadata
Sanitize line endings and remove blank lines between table rows
Standardize specification to 3.40 BPW in compliance with APEX-I-MiniPlus V2.1 architecture
Fix WikiText-2 perplexity to 5.1280 (+0.02 delta), 3.38 BPW, and 14.65 GB footprint across all sections
Fix bold markdown syntax on KL Divergence metric row
Fix raw LaTeX rendering in markdown table and callouts (use clean D_KL)
Align README with official V2.1 template, cross-links to standard edition, and empirical benchmarks
Add comprehensive model card and empirical benchmarks for Abliterated edition
Add companion mtp-Qwen3.8-35B-A3B-Distill-Q8_0.gguf speculative drafting module
Add companion mmproj-Q8_0.gguf multimodal vision projector
Add Qwen3.8-35B-A3B-Distill APEX-I-MiniPlus-V2.1 Abliterated GGUF (True Directional Refusal Ablation)
initial commit