Saif-1.1

An improved conversational assistant fine-tuned from Saif-1.0.

What's New in 1.1

• Trained on Claude Opus 4.6/4.7 generated data
• Faster response times vs Saif-1.0
• Better at coding, math, and reasoning
• Cleaner, more concise answers

Model Details

• Base model: Saif658/Saif-1.0
• Fine-tuned by: Saif658
• Training: QLoRA 4-bit, 500 steps
• Dataset: angrygiraffe/claude-opus-4.6-4.7-reasoning-8.7k
• License: Apache 2.0

Benchmarks vs Saif-1.0

Task	Saif-1.0	Saif-1.1
Prime check (Python)	✓ slower	✓ faster, better algo
Derivative (Math)	✓ 9.64s	✓ 3.87s
Factorial (JavaScript)	✓ 15.51s	✓ 7.72s

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("Saif658/Saif-1.1")
model = AutoModelForCausalLM.from_pretrained(
    "Saif658/Saif-1.1",
    torch_dtype=torch.float16,
    device_map="auto"
)

messages = [{"role": "user", "content": "your message here"}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")
outputs = model.generate(inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Limitations

Small 3B model — may struggle with very complex reasoning or long context tasks.