mistral-7b-threads-finetuned

Mistral 7B — Viral Twitter/X Threads (QLoRA Fine-tune)

A fine-tuned version of Mistral 7B Instruct v0.3, specialized in writing viral Twitter/X threads with hooks, storytelling, analogies, and CTAs.

Model Details

• Base model: mistralai/Mistral-7B-Instruct-v0.3
• Method: QLoRA (4-bit quantization + LoRA adapters)
• Trainable parameters: ~160M / 7.2B (~1.5%)
• LoRA config: r=16, alpha=32, dropout=0.05
• Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• Language: English
• License: Apache 2.0 (same as base model)

Training

Usage

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

# Load base model in 4-bit
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16,
)

base_model = AutoModelForCausalLM.from_pretrained(
    "mistralai/Mistral-7B-Instruct-v0.3",
    quantization_config=bnb_config,
    device_map="auto",
)

tokenizer = AutoTokenizer.from_pretrained("nyri/mistral-7b-threads-finetuned")

# Load LoRA adapters
model = PeftModel.from_pretrained(base_model, "nyri/mistral-7b-threads-finetuned")

# Generate
prompt = '<s>[INST] You are an expert at writing viral Twitter/X threads. Write a viral thread about the psychology of pricing. [/INST]'
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=300, temperature=0.7, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Intended Use

Generating engaging Twitter/X thread drafts. The model produces structured threads (hook → body → CTA) optimized for engagement.

Limitations

• Trained on English threads only
• Output quality varies by topic — works best on business, tech, psychology, and productivity
• Not a replacement for human editing — use as a first draft generator
• 500 training examples = style transfer, not deep domain knowledge

Developed by

Nylan Richard