Tag

Qwen3-VL-32B Ultra Uncensored Heretic — MiniMax-H3 ComfyUI INT8 ConvRot

Provides ComfyUI-ready INT8 MiniMax‑H3 checkpoints (conditioning encoder plus optional generation tail) for a Heretic-edited Qwen3‑VL‑32B source; preserves the vision tower in BF16 and uses row-wise ConvRot INT8 quantization to reduce VRAM needs for ~32GB GPUs. Not a full Transformers generation repository.

Qwen3-VL-32B Ultra Uncensored Heretic — H3 ComfyUI INT8 ConvRot

ComfyUI-ready H3 conditioning encoder builds for Qwen3-VL-32B: a BF16 full-precision checkpoint, an INT8 ConvRot quantized checkpoint, and an optional generation tail (layers 50–63). Retains vision tower in BF16 and targets H3 workflows and lower-VRAM systems.

stablediffusiontutorials/Minimax-H3

Installation-oriented dataset that packages ComfyUI-ready files and instructions for running MiniMax H3 locally — includes pruned/INT8/BF16 checkpoints, matching Qwen3-VL text encoders, video/audio VAEs, and official ComfyUI workflow templates for joint audio+video generation.

Beyond Simply Environment Scaling: Designing Effective Environment Distributions for Multimodal Agent Learning

Analyzes how to build effective training environment distributions for multimodal agents and proposes Ability-aware Environment Selection (AES) and Hierarchical Difficulty Curriculum (HDC) to improve diversity and difficulty scheduling, yielding large relative gains in experiments.

Qwen3-VL-32B Heretic (MiniMax-H3 text encoder) — NVFP4

An uncensored NVFP4-quantized text encoder for MiniMax-H3 video generation that fits on a single 16 GB GPU. Mixed-precision bake (mostly NVFP4, embedding left as INT8), preserves ConvRot rotation semantics, and includes the unrotate step required to avoid corrupted conditioning.

Qwen3.8-27B

A 27B-parameter causal language model with a native vision encoder for image/video+text understanding, long-horizon agentic tasks, and tunable thinking-mode reasoning. Native 262,144-token context (extensible to 1,000,000) and production-focused inference recipes.

AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning

Converts sparse trajectory-level rewards into turn-level credit by aggregating token-level teacher–student log-probability gaps and recursively updating a Bayesian belief in log-odds; produces turn-wise reweighting for policy optimization without an extra critic or rollouts.

MiniMax-H3-Prompt-Rewriter-LoRA

Turns a short prompt plus aspect ratio and duration into a structured, shot-by-shot audio-video description for text-to-audio-video generation. A PEFT LoRA on Qwen3.6-27B that expands timing, camera motion, continuity, and synchronized diegetic/non‑diegetic sound; text-only and requires MiniMax-H3 + LightX2V to produce final AV.

Qwen3.8-2.4T-A95B

A MoE causal large language model for long-horizon agents, coding, and multi-step reasoning: 2.4T parameters (95B activated), native 262,144-token context (extensible to 1,010,000), multi-token prediction, and configurable thinking-mode reasoning controls.

Qwen3.8-2.4T-A95B-FP8

FP8-quantized checkpoint of the Qwen3.8 text-only causal LLM (2.4T params, 95B activated) for text-generation; preserves near-original performance, supports very long contexts (262k–1M), Mixture-of-Experts architecture, and is compatible with vLLM/SGLang/TokenSpeed. Thinking mode and preserve_thinking are enabled by default.

Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA

A research report proposing a continual-learning agent workflow that pairs recursive self-improvement with a Mixture-of-LoRA design: freeze a foundation model, compose specialist LoRA adapters routed per user turn, and support them with long-context RL and post-training infrastructure.

Qwen Sharp Chat Templates

Provides a drop-in Jinja chat template for Qwen 3.5/3.6/3.8 that reduces reasoning-token waste, enforces a concise terseness system prompt, and preserves in-chat reasoning and tool-call rendering across turns. Terseness is on by default but switchable per request; no model weights are changed.