Solstice-AI/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU-MXFP4-1M

feat: update tokenizer_config chat_template to Solstice 10-level solo + swarm

feat: update to Solstice 10-level solo + swarm cognitive architecture

feat(tokenizer): update tokenizer_config.json for NEO-CODER support

fix(readme): remove Anvil section (wrong for safetensors repo), fix vLLM command

fix(config): change model_type from qwen3_5_text to qwen3_5, add missing text_config block

fix(mtp): restore complete 15-tensor MTP drafter from DavidAU base (BF16, unquantized) (#1)

fix(mtp): add index mapping model.safetensors and model-mtp-restored.safetensors

fix(mtp): add root num_nextn_predict_layers for vLLM speculative decoding

docs: update chat template with <think> reasoning tokens and tool calling schema

docs: note 4-bit nvfp4 KV cache option in vLLM

docs: update vLLM to 2026 serving flags (--kv-cache-dtype fp8, chunked prefill, prefix caching)

docs: 3x expanded model card with OCP MXFP4 microscaling specs, Anvil TurboQuant 1M recipes, and clean sweep benchmarks

docs: extensive marketing documentation, DavidAU co-branding, Opus 4.6 telemetry, and SEO tags

docs: extensive marketing documentation, DavidAU co-branding, Opus 4.6 telemetry, and SEO tags

docs: highlight verified benchmark dominance over Claude Opus 4.6 Max across SWE-bench Pro, OSWorld, and LiveCodeBench

docs: update exact power-of-two 1,048,576 (2^20) context telemetry and Anvil quickstart

fix: set exact power-of-two max_position_embeddings to 1,048,576 (2^20)

feat: add mmproj-BF16.gguf for native MTP and multimodal vision support

feat: add model-mtp-restored.safetensors for native MTP and multimodal vision support

feat: add video_preprocessor_config.json for native MTP and multimodal vision support

feat: add preprocessor_config.json for native MTP and multimodal vision support

Upload README.md with huggingface_hub

Upload README.md with huggingface_hub

feat: native 1M context release with baked-in YaRN scaling

initial commit