Hoodx

mira-agent-instinct

Deploy Dedicated

README

License: apache-2.0

Overview

Instinct mode: fast heuristic policy that reduces reasoning tokens while preserving first-attempt accuracy.
Agent layer: smolagents / Transformers Agent Toolkit compatible.
RSI: offline self-play fine-tuning with DPO, frozen improvement operator, lineage tracking.
Safety: gated repo, guardrails, audit log, explicit limitations.

Model Details

Base model: Qwen/Qwen3-8B
Architecture: decoder-only transformer
Context length: 32k
Training: supervised fine-tune on agent trajectories + self-play synthetic data
Quantized variants: Q4_K_M, Q8_0

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer model_id = "Hoodx/mira-agent-instinct" ## Loading ```pythonfrom transformers import AutoModelForCausalLM, AutoTokenizer model_id = "Hoodx/mira-agent-instinct"tokenizer = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

The LoRA adapter (adapter_model.safetensors, r=8, alpha=8) is also kept in the repo for reference. tokenizer = AutoTokenizer.from_pretrained(model_id) model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

messages = [ {"role": "user", "content": "Plan a 3-step research workflow and execute step 1"} ] inputs = tokenizer.apply_chat_template(messages, return_tensors="pt") outputs = model.generate(**inputs)

text

### smolagents ```pythonfrom smolagents import CodeAgent, HfApiModel model = HfApiModel(model_id="Hoodx/mira-agent-instinct")agent = CodeAgent(model=model, tools=[...])agent.run("...")

Demo

See the HuggingFace Space: Hoodx/mira-agent-instinct-demo

Evaluation

Results in .eval_results/:

Terminal-Bench 2.1
SWE-Bench Verified
AgentBench

Evaluation

Table with columns: Metric, Value, Notes
Metric	Value	Notes
Perplexity (held-out)	1.24	10% held-out split of `Hoodx/mira-dataset`
Task completion rate	TBD	50 agent prompts, human-rated
Training loss (final)	0.2154	From eval on 1025 examples

Eval script: eval_mira.py in slab1/mira-training-scripts (run on Colab/Kaggle).

Limitations

Not AGI. Agent-oriented LLM with heuristic compression.
Recursive self-improvement is offline and eval-gated.
Tool use requires guardrails.

Citation

bibtex

@misc{mira-agent-instinct,  title={Mira AI Agent Model – Instinctive AGI / RSI},  author={Hoodx},  year={2026}}

License

Apache-2.0

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

Hoodx

Model Tree

Base

Qwen/Qwen3-8B

Fine-tuned

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Overview

Instinct mode: fast heuristic policy that reduces reasoning tokens while preserving first-attempt accuracy.
Agent layer: smolagents / Transformers Agent Toolkit compatible.
RSI: offline self-play fine-tuning with DPO, frozen improvement operator, lineage tracking.
Safety: gated repo, guardrails, audit log, explicit limitations.

Model Details

Base model: Qwen/Qwen3-8B
Architecture: decoder-only transformer
Context length: 32k
Training: supervised fine-tune on agent trajectories + self-play synthetic data
Quantized variants: Q4_K_M, Q8_0

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer model_id = "Hoodx/mira-agent-instinct" ## Loading ```pythonfrom transformers import AutoModelForCausalLM, AutoTokenizer model_id = "Hoodx/mira-agent-instinct"tokenizer = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

from transformers import AutoModelForCausalLM, AutoTokenizer model_id = "Hoodx/mira-agent-instinct" ## Loading ```pythonfrom transformers import AutoModelForCausalLM, AutoTokenizer model_id = "Hoodx/mira-agent-instinct"tokenizer = AutoTokenizer.from_pretrained(model_id)model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

text

### smolagents```pythonfrom smolagents import CodeAgent, HfApiModelmodel = HfApiModel(model_id="Hoodx/mira-agent-instinct")agent = CodeAgent(model=model, tools=[...])agent.run("...")

### smolagents ```pythonfrom smolagents import CodeAgent, HfApiModel model = HfApiModel(model_id="Hoodx/mira-agent-instinct")agent = CodeAgent(model=model, tools=[...])agent.run("...")

Demo

See the HuggingFace Space: Hoodx/mira-agent-instinct-demo

Evaluation

Results in .eval_results/:

Terminal-Bench 2.1
SWE-Bench Verified
AgentBench

Evaluation

Table with columns: Metric, Value, Notes
Metric	Value	Notes
Perplexity (held-out)	1.24	10% held-out split of `Hoodx/mira-dataset`
Task completion rate	TBD	50 agent prompts, human-rated
Training loss (final)	0.2154	From eval on 1025 examples

Eval script: eval_mira.py in slab1/mira-training-scripts (run on Colab/Kaggle).

Limitations

Not AGI. Agent-oriented LLM with heuristic compression.
Recursive self-improvement is offline and eval-gated.
Tool use requires guardrails.

Citation

bibtex

@misc{mira-agent-instinct,  title={Mira AI Agent Model – Instinctive AGI / RSI},  author={Hoodx},  year={2026}}

@misc{mira-agent-instinct,  title={Mira AI Agent Model – Instinctive AGI / RSI},  author={Hoodx},  year={2026}}

License

Apache-2.0