Views
No views yet
| Attribute | Value |
|---|---|
| Developer | Antagon Inc. (CAGE: 17E75, UEI: KBSGT7CZ4AH3) |
| Base Model | Qwen/Qwen2-7B-Instruct |
| Method | LoRA (Low-Rank Adaptation) |
| Trainable Parameters | 40.4M (0.53% of 7.6B total) |
| Training Data | 11.7M critique examples |
| Training Hardware | NVIDIA H100 PCIe (80GB) via Lambda Labs GPU Grant |
| License | Apache 2.0 |
1learning_rate: 2e-4
2lr_scheduler: cosine
3warmup_steps: 500
4batch_size: 32 (effective)
5max_sequence_length: 512
6epochs: 1
7lora_r: 16
8lora_alpha: 32
9lora_dropout: 0.05
10target_modules: [q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj]pip install transformers peft torch1from transformers import AutoTokenizer, AutoModelForCausalLM
2from peft import PeftModel
3
4# Load base model
5base_model = AutoModelForCausalLM.from_pretrained(
6 "Qwen/Qwen2-7B-Instruct",
7 torch_dtype=torch.bfloat16,
8 device_map="auto"
9)
10
11# Load tokenizer
12tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-7B-Instruct")
13
14# Load LoRA adapter
15model = PeftModel.from_pretrained(base_model, "Antagon/MiniCrit-7B")1def critique_rationale(rationale: str) -> str:
2 prompt = f"### Rationale:\n{rationale}\n\n### Critique:\n"
3
4 inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
5 outputs = model.generate(
6 **inputs,
7 max_new_tokens=256,
8 temperature=0.7,
9 do_sample=True,
10 pad_token_id=tokenizer.eos_token_id
11 )
12
13 response = tokenizer.decode(outputs[0], skip_special_tokens=True)
14 return response.split("### Critique:\n")[-1]
15
16# Example
17rationale = "AAPL long: MACD bullish crossover with supporting momentum."
18critique = critique_rationale(rationale)
19print(critique)Input: "META long: Bollinger Band expansion with supporting momentum."
Output: "While Bollinger Band expansion can signal volatility, META's recent
expansion isn't necessarily predictive; it could be a reaction to news, not
a precursor to sustained movement. Furthermore, relying solely on momentum
without considering overbought/oversold levels may lead to premature entry,
especially if the expansion is already near its peak."| Metric | Value |
|---|---|
| Initial Loss | 1.8573 |
| Final Loss | 0.7869 |
| Loss Reduction | 57.6% |
| Gradient Norm (avg) | 0.45 |
1@misc{minicrit7b2026,
2 title={MiniCrit-7B: Adversarial AI Critique for Trading Signal Validation},
3 author={Ousley, William Alexander and Ousley, Jacqueline Villamor},
4 year={2026},
5 publisher={Antagon Inc.},
6 url={https://huggingface.co/Antagon/MiniCrit-7B}
7}