Downloads · 30 days
10
23% of all-time downloads
Papajams/orbura-dataviz-mistral-7b-autoscientist
orbura-dataviz-mistral-7b-autoscientist is a text generation model from Papajams. Use it when you need the model to write or continue text. It is set up for peft.
- Base model: mistralai/Mistral-7B-Instruct-v0.2 - Fine-tuning method: LoRA (r=16, alpha=32) via AutoScientist co-optimization - Dataset: Papajams/orbura-dataviz-augmented (5,000 rows with reasoning traces) - Task: Da…
Downloads · 30 days
10
23% of all-time downloads
All-time downloads
43
Public
Repo size
54.6 MB
Likes
0
Public
Click a slice to open those files.
.safetensors54.6 MB · 94%
From the Hugging Face model README
Naive baseline: 17.6% on 2k validation set.
Run the eval harness:
cd showcases/autoscientist-dataviz
python3 eval_finetuned.py --baseline-only
Scoring metrics (per-task deterministic evaluation):
| Task | Scoring method |
|---|---|
chart_qa | Exact or numeric-tolerance match |
chart_to_code / data_to_code | Chart function + data value overlap (≥80%) |
code_to_desc | Chart type + ≥2 numeric mentions |
style_transfer | Normalized code equivalence |
fix_code | Normalized code equivalence |
chart_choice | First-word (chart type) match |
This is a LoRA adapter. Load it with PEFT:
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
model = PeftModel.from_pretrained(base, "Papajams/orbura-dataviz-mistral-7b-autoscientist")
tokenizer = AutoTokenizer.from_pretrained("Papajams/orbura-dataviz-mistral-7b-autoscientist")
messages = [{"role": "user", "content": "Answer the question using the chart data.\n\n- A: 45\n- B: 78"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(output[0], skip_special_tokens=True))
# 1. Generate the dataset
cd showcases/autoscientist-dataviz
python3 generate_full.py
python3 prepare_adaption.py
# 2. Run Adaption augmentation
python3 run_adaption.py --input data/adaption_train_5k.jsonl
# 3. Train via AutoScientist (adaptionlabs.ai/auto-scientist)
# Upload the augmented dataset, select Mistral-7B, run co-optimization
# 4. Evaluate
python3 eval_finetuned.py --baseline-only
# 5. Publish
python3 publish_hf.py --model path/to/model
Apache 2.0 (base model license applies). Dataset and model generated for the AutoScientist Challenge by Adaption Labs.