Charley890/adaption_financial_ticker_completions

original_model_name: "meta-llama/Llama-3.3-70B-Instruct-Reference"

model_name: "adaption_financial_ticker_completions"

model_type: "LoRA adapter"

domain: "Market Analysis"

description: > A finance-focused language model adaptation designed for market analysis, financial reasoning, market trends, price interpretation, risk analysis, and structured financial question answering.

credit: Adaptive_data by "Adaption Labs" platform: "Adaption Labs" workflow: "Adaption Labs AutoScientist" challenge: "AutoScientist Challenge"

training: method: "SFT" adaptation: "LoRA" data_format: "chat"

hyperparameters: lora: true lora_r: 64 lora_alpha: 128 lora_dropout: 0 epochs: 3 n_evals: 5 batch_size: "max" learning_rate: 0.00005 min_lr_ratio: 0.1 warmup_ratio: 0.05 weight_decay: 0.02 max_grad_norm: 1

scheduler: type: "cosine" num_cycles: 0.5

target_modules: "all-linear"

mathematical_market_analysis:

percentage_change: formula: "((P_current - P_previous) / P_previous) * 100" purpose: "Measures percentage price movement."

return: formula: "(P_1 - P_0) / P_0" purpose: "Measures asset return between two prices."

moving_average: formula: "MA_n = (P_1 + P_2 + ... + P_n) / n" purpose: "Identifies the underlying market trend."

volatility: formula: "sigma = sqrt(Var(R))" purpose: "Measures variability in market returns."

sharpe_ratio: formula: "Sharpe = (R_p - R_f) / sigma_p" purpose: "Measures risk-adjusted return."

evaluation:

base_model: quality_score: 2.0 overall_win_rate: "37%" market_analysis_win_rate: "25%" grade: "E"

adapted_model: quality_score: 8.6 overall_win_rate: "63%" market_analysis_win_rate: "75%" grade: "B"

improvement: quality: from: 2.0 to: 8.6 relative_improvement: "+330%"

overall_win_rate:
  from: "37%"
  to: "63%"
  improvement: "+26 percentage points"

market_analysis_win_rate:
  from: "25%"
  to: "75%"
  improvement: "+50 percentage points"

results: summary: > The adapted model demonstrates substantial improvement in market-analysis reasoning, financial interpretation, and structured responses compared with the base model.

strongest_result: metric: "Market Analysis Win Rate" before: "25%" after: "75%" improvement: "+50 percentage points"

capabilities:

• "Market trend analysis"
• "Price movement interpretation"
• "Financial reasoning"
• "Technical-analysis explanation"
• "Risk and return analysis"
• "Market scenario analysis"
• "Financial question answering"

example_use: prompt: > Analyze an asset that increased from $100 to $125 and explain the percentage movement and possible market interpretation.

calculation: formula: "((125 - 100) / 100) * 100" result: "25%"

limitations:

• "Market analysis is not financial advice."
• "Predictions are not guaranteed."
• "Market conditions can change rapidly."
• "Financial outputs should be independently verified."

provenance: data_source: "Adaptive Data" enhancement_platform: "Adaption Labs" training_platform: "Adaption Labs AutoScientist" original_model: "meta-llama/Llama-3.3-70B-Instruct-Reference"

---

Model Training

A LORA adapter for meta-llama/Llama-3.3-70B-Instruct-Reference. This model was trained with SFT using Adaption's AutoScientist on the financial_ticker_completions dataset.

AutoScientist Config

{
  "job_id": "dc54b8ee-f0fc-44a7-b0a6-8dfaf77592da",
  "training_experiment_id": "81e39c69-456b-4347-bef7-f615a058b5a1",
  "original_model_name": "meta-llama/Llama-3.3-70B-Instruct-Reference",
  "trained_model_name": "adaption_financial_ticker_completions",
  "training_method": "sft",
  "training_type": "lora",
  "data_format": "chat",
  "hyperparams": {
    "lora": "true",
    "lora_r": 64,
    "n_evals": 5,
    "n_epochs": 3,
    "batch_size": "max",
    "lora_alpha": 128,
    "lora_dropout": 0,
    "min_lr_ratio": 0.1,
    "warmup_ratio": 0.05,
    "weight_decay": 0.02,
    "learning_rate": 0.00005,
    "max_grad_norm": 1,
    "base_model_size": "70B",
    "train_on_inputs": "false",
    "training_method": "sft",
    "lr_scheduler_type": "cosine",
    "scheduler_num_cycles": 0.5,
    "lora_trainable_modules": "all-linear"
  }
}

Training Data

The model was trained on 1,473 rows of adapted data with the following domain distribution: market-analysis (61%), math (21%), code (6%), news (6%), science (3%), personal-finance (2%), corporate-business (1%), legal (0%), transportation (0%), games (0%), other (0%), technology (0%), data-analysis-visualization (0%).

Model Evaluation

The model was evaluated on an in-distribution held-out test set as well as a broader domain-specific test set to measure generalization.

Domain	Win rate vs. base model
market-analysis	75%

How to use

pip install torch transformers peft

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

BASE = "meta-llama/Llama-3.3-70B-Instruct-Reference"
ADAPTER = "<this-repo-id>"

device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float32 if device == "cpu" else torch.bfloat16

base = AutoModelForCausalLM.from_pretrained(BASE, dtype=dtype).to(device)
model = PeftModel.from_pretrained(base, ADAPTER)
# Optional: merge the LoRA weights into the base for faster inference
model = model.merge_and_unload()
model.eval()

tokenizer = AutoTokenizer.from_pretrained(BASE)
messages = [{"role": "user", "content": "Hello!"}]
text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(device)

with torch.inference_mode():
    out = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))