google/gemma-4-31B
Image-Text-to-Text ā¢ 33B ā¢ Updated ā¢ 184k ā¢ 283
Samir ARISS
samariss
AI & ML interests
None yet
Recent Activity
liked a model 1 day ago
google/gemma-4-31B liked a model 1 day ago
google/gemma-4-31B-it liked a model 2 days ago
meta-llama/Llama-Guard-3-11B-VisionOrganizations
None yet
liked 2
models 1 day ago
liked a
model 2 days ago
liked a
model 4 days ago
liked a
model 10 days ago
liked a
model 17 days ago
liked a
model 19 days ago
reactedto SeaWolf-AI's post with š„ about 1 month ago
Post
8180
š Introducing MARL ā Runtime Middleware That Reduces LLM Hallucination Without Fine-Tuning
Now available on PyPI Ā· GitHub Ā· ClawHub Ā· HuggingFace
AI models sense they could be wrong, but they can't actually fix what's broken.
š¤ Live A/B test: VIDraft/MARL
We evaluated 9 SOTA models (GPT-5.2, Claude Opus 4.6, Gemini 3 Pro, etc.) across 1,800 assessments in FINAL Bench and found a 39.2%p gap between "recognizing potential errors (MA=0.694)" and "actually finding and fixing them (ER=0.302)."
MARL (Model-Agnostic Runtime Middleware for LLMs) was built to close this metacognitive gap. It decomposes a single LLM call into a 5-stage expert pipeline (Hypothesis ā Solver ā Auditor ā Adversarial Verifier ā Synthesizer), transforming "answer in one shot" into "think, doubt, correct, and rewrite."
No weight modification ā works instantly with GPT-5.4, Claude, Gemini, Llama, or any OpenAI API-compatible LLM by changing one line: base_url. Ships with 9 domain-specific emergence engines (invention, pharma, genomics, chemistry, ecology, law, and more ā 5,538 expert data items) activated by a simple tag like model="gpt-5.4::pharma".
pip install marl-middleware
MARL is also officially registered on ClawHub, the skill marketplace of OpenClaw ā an AI agent platform with 260K+ developers and 3,200+ skills. It's the first middleware in the Reasoning Enhancement category. One command ā clawhub install marl-middleware ā gives your AI agent a metacognition upgrade.
š Technical deep dive: https://huggingface.co/blog/FINAL-Bench/marl-middleware
š¦ PyPI: https://pypi.org/project/marl-middleware/
š GitHub: https://github.com/Vidraft/MARL
š¦ ClawHub: https://clawhub.ai/Cutechicken99/marl-middleware
#MARL #LLM #Hallucination #Metacognition #MultiAgent #AIMiddleware #FINALBench #OpenClaw #ClawHub #PyPI #AGI #HuggingFace #ReasoningAI #SelfCorrection #GlassBoxAI
Now available on PyPI Ā· GitHub Ā· ClawHub Ā· HuggingFace
AI models sense they could be wrong, but they can't actually fix what's broken.
š¤ Live A/B test: VIDraft/MARL
We evaluated 9 SOTA models (GPT-5.2, Claude Opus 4.6, Gemini 3 Pro, etc.) across 1,800 assessments in FINAL Bench and found a 39.2%p gap between "recognizing potential errors (MA=0.694)" and "actually finding and fixing them (ER=0.302)."
MARL (Model-Agnostic Runtime Middleware for LLMs) was built to close this metacognitive gap. It decomposes a single LLM call into a 5-stage expert pipeline (Hypothesis ā Solver ā Auditor ā Adversarial Verifier ā Synthesizer), transforming "answer in one shot" into "think, doubt, correct, and rewrite."
No weight modification ā works instantly with GPT-5.4, Claude, Gemini, Llama, or any OpenAI API-compatible LLM by changing one line: base_url. Ships with 9 domain-specific emergence engines (invention, pharma, genomics, chemistry, ecology, law, and more ā 5,538 expert data items) activated by a simple tag like model="gpt-5.4::pharma".
pip install marl-middleware
MARL is also officially registered on ClawHub, the skill marketplace of OpenClaw ā an AI agent platform with 260K+ developers and 3,200+ skills. It's the first middleware in the Reasoning Enhancement category. One command ā clawhub install marl-middleware ā gives your AI agent a metacognition upgrade.
š Technical deep dive: https://huggingface.co/blog/FINAL-Bench/marl-middleware
š¦ PyPI: https://pypi.org/project/marl-middleware/
š GitHub: https://github.com/Vidraft/MARL
š¦ ClawHub: https://clawhub.ai/Cutechicken99/marl-middleware
#MARL #LLM #Hallucination #Metacognition #MultiAgent #AIMiddleware #FINALBench #OpenClaw #ClawHub #PyPI #AGI #HuggingFace #ReasoningAI #SelfCorrection #GlassBoxAI
liked 7
models about 1 month ago
zai-org/GLM-5-FP8
Text Generation ā¢ 754B ā¢ Updated ā¢ 1.61M ā¢ 171
Qwen/Qwen3.5-9B
Image-Text-to-Text ā¢ 10B ā¢ Updated ā¢ 5.4M ā¢ ā¢ 1.24k
Qwen/Qwen3.5-35B-A3B-FP8
Image-Text-to-Text ā¢ 36B ā¢ Updated ā¢ 1.98M ā¢ 143
Qwen/Qwen3.5-35B-A3B
Image-Text-to-Text ā¢ 36B ā¢ Updated ā¢ 3.56M ā¢ ā¢ 1.36k
Qwen/Qwen3.5-397B-A17B-FP8
Image-Text-to-Text ā¢ 403B ā¢ Updated ā¢ 1.07M ā¢ 151
Qwen/Qwen3.5-4B
Image-Text-to-Text ā¢ 5B ā¢ Updated ā¢ 2.94M ā¢ 447
nvidia/llama-nemotron-embed-vl-1b-v2
Sentence Similarity ā¢ 2B ā¢ Updated ā¢ 83.1k ā¢ 57
liked 4
models about 2 months ago
intfloat/e5-mistral-7b-instruct
Feature Extraction ā¢ 7B ā¢ Updated ā¢ 362k ā¢ ā¢ 563
Qwen/Qwen3.5-397B-A17B
Image-Text-to-Text ā¢ 403B ā¢ Updated ā¢ 807k ā¢ ā¢ 1.43k
zai-org/GLM-5
Text Generation ā¢ 754B ā¢ Updated ā¢ 491k ā¢ ā¢ 1.98k
MiniMaxAI/MiniMax-M2.1
Text Generation ā¢ 229B ā¢ Updated ā¢ 34.4k ā¢ ā¢ 1.27k
liked a
model 2 months ago