Instructions to use Agnuxo/CAJAL-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Agnuxo/CAJAL-4B with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Agnuxo/CAJAL-4B",
	filename="CAJAL-4B-f16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use Agnuxo/CAJAL-4B with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Agnuxo/CAJAL-4B:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Agnuxo/CAJAL-4B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Agnuxo/CAJAL-4B:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf Agnuxo/CAJAL-4B:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Agnuxo/CAJAL-4B:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Agnuxo/CAJAL-4B:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Agnuxo/CAJAL-4B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Agnuxo/CAJAL-4B:Q4_K_M

Use Docker

docker model run hf.co/Agnuxo/CAJAL-4B:Q4_K_M

LM Studio
Jan

vLLM

How to use Agnuxo/CAJAL-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Agnuxo/CAJAL-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Agnuxo/CAJAL-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Agnuxo/CAJAL-4B:Q4_K_M

Ollama
How to use Agnuxo/CAJAL-4B with Ollama:
```
ollama run hf.co/Agnuxo/CAJAL-4B:Q4_K_M
```

Unsloth Studio new

How to use Agnuxo/CAJAL-4B with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Agnuxo/CAJAL-4B to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Agnuxo/CAJAL-4B to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Agnuxo/CAJAL-4B to start chatting

Pi new

How to use Agnuxo/CAJAL-4B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Agnuxo/CAJAL-4B:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "CAJAL-4B"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Agnuxo/CAJAL-4B with Docker Model Runner:
```
docker model run hf.co/Agnuxo/CAJAL-4B:Q4_K_M
```

Lemonade

How to use Agnuxo/CAJAL-4B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Agnuxo/CAJAL-4B:Q4_K_M

Run and chat with the model

lemonade run user.CAJAL-4B-Q4_K_M

List all available models

lemonade list

Agnuxo commited on 4 days ago

Commit

82cbcff

verified ·

1 Parent(s): e2c341c

Add harness_best.json

Browse files

Files changed (1) hide show

harness_best.json +496 -0

harness_best.json ADDED Viewed

	@@ -0,0 +1,496 @@

+{
+  "best_score": 6.2,
+  "best_run": 78,
+  "best_result": {
+    "run_id": 78,
+    "model": "cajal-4b-f16",
+    "topic": "zkSNARK-Proven Correctness of Leader Rotation in Permissionless Consensus",
+    "score": 6.2,
+    "scores": {
+      "sections": {
+        "abstract": 5.5,
+        "introduction": 5.1,
+        "methodology": 4.1,
+        "results": 3.3,
+        "discussion": 3.6,
+        "conclusion": 4.3,
+        "references": 5.1
+      },
+      "overall": 6.2,
+      "novelty": 5.6,
+      "reproducibility": 5,
+      "citation_quality": 5.3,
+      "judges": [
+        "Cerebras-Qwen235B",
+        "Cerebras-Llama8B",
+        "Mistral",
+        "Sarvam",
+        "NVIDIA",
+        "Inception-Mercury2",
+        "Cohere-CommandA",
+        "Cloudflare-Qwen3",
+        "NVIDIA-Devstral",
+        "Cohere-Command-A",
+        "Cohere-R7B",
+        "Mistral-Medium",
+        "Mistral-Large",
+        "Sarvam-KeyVariant-2",
+        "Sarvam-KeyVariant-3",
+        "OpenRouter-GPT-OSS-Free",
+        "Cohere-Aya-Expanse",
+        "Inception-Mercury2-Key2",
+        "Cerebras-Qwen235B-Key2"
+      ],
+      "judge_count": 19,
+      "judge_details": [
+        {
+          "judge": "Cerebras-Qwen235B",
+          "scores": {
+            "abstract": 7,
+            "introduction": 7,
+            "methodology": 5,
+            "results": 4,
+            "discussion": 5,
+            "conclusion": 6,
+            "references": 3,
+            "novelty": 5,
+            "reproducibility": 4,
+            "citation_quality": 3
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Cerebras-Llama8B",
+          "scores": {
+            "abstract": 8,
+            "introduction": 8,
+            "methodology": 6,
+            "results": 7,
+            "discussion": 6,
+            "conclusion": 7,
+            "references": 9,
+            "novelty": 8,
+            "reproducibility": 8,
+            "citation_quality": 9
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Mistral",
+          "scores": {
+            "abstract": 7,
+            "introduction": 6,
+            "methodology": 4,
+            "results": 4,
+            "discussion": 5,
+            "conclusion": 5,
+            "references": 5,
+            "novelty": 6,
+            "reproducibility": 3,
+            "citation_quality": 4
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Sarvam",
+          "scores": {
+            "abstract": 7,
+            "introduction": 6,
+            "methodology": 5,
+            "results": 4,
+            "discussion": 5,
+            "conclusion": 7,
+            "references": 4,
+            "novelty": 7,
+            "reproducibility": 3,
+            "citation_quality": 5
+          },
+          "feedback": null
+        },
+        {
+          "judge": "NVIDIA",
+          "scores": {
+            "abstract": 8,
+            "introduction": 9,
+            "methodology": 8,
+            "results": 7,
+            "discussion": 8,
+            "conclusion": 8,
+            "references": 9,
+            "novelty": 7,
+            "reproducibility": 8,
+            "citation_quality": 9
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Inception-Mercury2",
+          "scores": {
+            "abstract": 6,
+            "introduction": 6,
+            "methodology": 5,
+            "results": 4,
+            "discussion": 5,
+            "conclusion": 6,
+            "references": 4,
+            "novelty": 4,
+            "reproducibility": 3,
+            "citation_quality": 4
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Cohere-CommandA",
+          "scores": {
+            "abstract": 8,
+            "introduction": 9,
+            "methodology": 8,
+            "results": 4,
+            "discussion": 7,
+            "conclusion": 7,
+            "references": 7,
+            "novelty": 7,
+            "reproducibility": 9,
+            "citation_quality": 8
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Cloudflare-Qwen3",
+          "scores": {
+            "abstract": 6,
+            "introduction": 5,
+            "methodology": 5,
+            "results": 4,
+            "discussion": 5,
+            "conclusion": 5,
+            "references": 5,
+            "novelty": 5,
+            "reproducibility": 5,
+            "citation_quality": 4
+          },
+          "feedback": null
+        },
+        {
+          "judge": "NVIDIA-Devstral",
+          "scores": {
+            "abstract": 8,
+            "introduction": 7,
+            "methodology": 6,
+            "results": 5,
+            "discussion": 5,
+            "conclusion": 6,
+            "references": 6,
+            "novelty": 7,
+            "reproducibility": 7,
+            "citation_quality": 6
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Cohere-Command-A",
+          "scores": {
+            "abstract": 8,
+            "introduction": 7,
+            "methodology": 6,
+            "results": 5,
+            "discussion": 6,
+            "conclusion": 7,
+            "references": 7,
+            "novelty": 7,
+            "reproducibility": 6,
+            "citation_quality": 8
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Cohere-R7B",
+          "scores": {
+            "abstract": 8,
+            "introduction": 8,
+            "methodology": 7,
+            "results": 7,
+            "discussion": 7,
+            "conclusion": 7,
+            "references": 8,
+            "novelty": 8,
+            "reproducibility": 8,
+            "citation_quality": 8
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Mistral-Medium",
+          "scores": {
+            "abstract": 9,
+            "introduction": 8,
+            "methodology": 6,
+            "results": 5,
+            "discussion": 4,
+            "conclusion": 5,
+            "references": 5,
+            "novelty": 7,
+            "reproducibility": 6,
+            "citation_quality": 6
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Mistral-Large",
+          "scores": {
+            "abstract": 7,
+            "introduction": 6,
+            "methodology": 5,
+            "results": 4,
+            "discussion": 3,
+            "conclusion": 5,
+            "references": 6,
+            "novelty": 6,
+            "reproducibility": 5,
+            "citation_quality": 7
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Sarvam-KeyVariant-2",
+          "scores": {
+            "abstract": 7,
+            "introduction": 6,
+            "methodology": 4,
+            "results": 3,
+            "discussion": 3,
+            "conclusion": 5,
+            "references": 1,
+            "novelty": 6,
+            "reproducibility": 3,
+            "citation_quality": 1
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Sarvam-KeyVariant-3",
+          "scores": {
+            "abstract": 7,
+            "introduction": 6,
+            "methodology": 5,
+            "results": 4,
+            "discussion": 5,
+            "conclusion": 7,
+            "references": 4,
+            "novelty": 7,
+            "reproducibility": 3,
+            "citation_quality": 5
+          },
+          "feedback": null
+        },
+        {
+          "judge": "OpenRouter-GPT-OSS-Free",
+          "scores": {
+            "abstract": 5,
+            "introduction": 5,
+            "methodology": 3,
+            "results": 3,
+            "discussion": 3,
+            "conclusion": 4,
+            "references": 5,
+            "novelty": 4,
+            "reproducibility": 3,
+            "citation_quality": 5
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Cohere-Aya-Expanse",
+          "scores": {
+            "abstract": 9,
+            "introduction": 8,
+            "methodology": 7,
+            "results": 6,
+            "discussion": 7,
+            "conclusion": 8,
+            "references": 9,
+            "novelty": 6,
+            "reproducibility": 7,
+            "citation_quality": 9
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Inception-Mercury2-Key2",
+          "scores": {
+            "abstract": 7,
+            "introduction": 6,
+            "methodology": 5,
+            "results": 3,
+            "discussion": 0,
+            "conclusion": 0,
+            "references": 7,
+            "novelty": 5,
+            "reproducibility": 3,
+            "citation_quality": 6
+          },
+          "feedback": null
+        },
+        {
+          "judge": "Cerebras-Qwen235B-Key2",
+          "scores": {
+            "abstract": 7,
+            "introduction": 7,
+            "methodology": 5,
+            "results": 4,
+            "discussion": 5,
+            "conclusion": 6,
+            "references": 3,
+            "novelty": 5,
+            "reproducibility": 4,
+            "citation_quality": 3
+          },
+          "feedback": null
+        }
+      ],
+      "consensus": {
+        "abstract": 0.8,
+        "introduction": 0.76,
+        "methodology": 0.75,
+        "results": 0.75,
+        "discussion": 0.64,
+        "conclusion": 0.65,
+        "references": 0.56,
+        "novelty": 0.76,
+        "reproducibility": 0.58,
+        "citation_quality": 0.55
+      },
+      "overall_consensus": 0.68,
+      "feedback": null,
+      "scored_at": "2026-05-07T13:18:06.650Z",
+      "paper_type": "research",
+      "calibration": {
+        "field": "cs-distributed",
+        "field_confidence": 1,
+        "signals_summary": {
+          "word_count": 5971,
+          "sections_present": 7,
+          "sections_missing": [],
+          "red_flags": [
+            "low_vocabulary_diversity",
+            "excessive_repetition_ratio_0.277"
+          ],
+          "red_flag_count": 2,
+          "has_formal_proofs": true,
+          "has_equations": true,
+          "has_code": true,
+          "unique_refs": 17,
+          "has_placeholder_refs": false,
+          "depth_score": 1,
+          "evidence_markers": 11,
+          "deception_count": 0,
+          "deception_matches": [],
+          "grammar": {
+            "vocabulary_diversity": 0.234,
+            "is_monotone": false,
+            "is_low_vocabulary": true
+          },
+          "repetition_ratio": 0.277,
+          "code_quality": {
+            "blocks": 6,
+            "has_real_code": false,
+            "has_python": true
+          },
+          "math_formulas": 11,
+          "lean4": "none",
+          "tables": 0
+        },
+        "adjustments": {
+          "abstract": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)",
+            "llm_inflation_correction: 4.3 -> 4"
+          ],
+          "introduction": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)",
+            "llm_inflation_correction: 3.8 -> 3.6"
+          ],
+          "methodology": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)",
+            "llm_inflation_correction: 2.5 -> 2.6"
+          ],
+          "results": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)",
+            "llm_inflation_correction: 1.6 -> 1.8"
+          ],
+          "discussion": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)",
+            "llm_inflation_correction: 1.9 -> 2.1"
+          ],
+          "conclusion": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)"
+          ],
+          "references": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)"
+          ],
+          "novelty": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)",
+            "llm_inflation_correction: 3.2 -> 3.1"
+          ],
+          "reproducibility": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)",
+            "llm_inflation_correction: 2.2 -> 2.3"
+          ],
+          "citation_quality": [
+            "red_flag_penalty: -3 (low_vocabulary_diversity, excessive_repetition_ratio_0.277, code_blocks_are_template_not_real)"
+          ]
+        },
+        "adjustment_count": 10,
+        "reference_papers": [
+          "The Byzantine Generals Problem",
+          "Bitcoin: A Peer-to-Peer Electronic Cash System",
+          "In Search of an Understandable Consensus Algorithm"
+        ],
+        "false_positive_corrected": "code_blocks_are_template_not_real (live verification confirmed code executes)"
+      },
+      "live_verification": {
+        "verification_time_ms": 18217,
+        "citations": {
+          "total": 8,
+          "verified": 7,
+          "verification_rate": 88
+        },
+        "novelty": {
+          "searched": true,
+          "total_found": 5,
+          "novelty_concern": "low",
+          "max_similarity": 33
+        },
+        "code_execution": {
+          "total": 5,
+          "passed": 3,
+          "failed": 0
+        },
+        "lean4": {
+          "blocks_found": 0,
+          "verified": 0,
+          "has_unsubstantiated_claim": true
+        },
+        "adjustments": {
+          "reproducibility": "claims_formal_verification_without_lean4_code: cap at 3",
+          "reproducibility_cap": 3
+        },
+        "bonuses": {
+          "references": "crossref_verified_7/8(88%): +1 bonus",
+          "references_bonus": 1,
+          "citation_quality": "crossref_high_rate: +1 bonus",
+          "citation_quality_bonus": 1,
+          "novelty": "arxiv_no_similar_papers: +1 novelty bonus",
+          "novelty_bonus": 1,
+          "reproducibility": "code_executed_3/5_passed: +2 reproducibility bonus",
+          "reproducibility_bonus": 2,
+          "execution_proof_bonus": 1.5,
+          "execution_proof_note": "3 code block(s) executed successfully: +1.5 overall bonus (capped at 1.5)"
+        }
+      }
+    },
+    "words": 5915,
+    "paper_id": "paper-1778159681193",
+    "ts": "2026-05-07T15:18:12.388195"
+  },
+  "ts": "2026-05-07T15:18:12.389185"
+}