SRafi007/qwen3.5-0.8b-lora-lead-qualifier

Qwen3.5-0.8B Finetuned for B2B Sales Lead Extraction

This model is a fine-tuned version of unsloth/Qwen3.5-0.8B optimized for extracting structured information from B2B sales messages. It identifies lead types, authority levels, intent, and recommended actions.

Model Details

• —Developed by: SRafi007
• —Model type: Causal Language Model (LoRA Adapters)
• —Language(s): English
• —Base Model: Qwen/Qwen3.5-0.8B
• —Finetuning Tool: Unsloth

Training Configuration

• —Method: LoRA (Low-Rank Adaptation)
• —Rank (r): 16
• —Alpha: 32
• —Target Modules: qproj, kproj, vproj, oproj, gateproj, upproj, down_proj
• —Batch Size: 4
• —Gradient Accumulation: 4
• —Learning Rate: 2e-4
• —Epochs: 3.0
• —Precision: Float32 (Unsloth auto-fallback)
• —Optimizer: AdamW 8-bit

Training Results

• —Final Loss: 0.6111
• —Global Steps: 213
• —Training Runtime: ~14.8 minutes (on Tesla T4)

How to Use

from unsloth import FastLanguageModel
import torch

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "SRafi007/qwen3.5-0.8b-lora",
    max_seq_length = 512,
    load_in_4bit = True,
)
FastLanguageModel.for_inference(model)

messages = [
    {"role": "system", "content": "You are a B2B sales assistant. Analyze the user message and extract structured lead information."},
    {"role": "user", "content": "We are a startup exploring pricing and integration with Slack."},
]

inputs = tokenizer.apply_chat_template(messages, tokenize = True, add_generation_prompt = True, return_tensors = "pt").to("cuda")
outputs = model.generate(input_ids = inputs, max_new_tokens = 150)
print(tokenizer.decode(outputs[0], skip_special_tokens = True))