File size: 2,404 Bytes
78aff1a 13d4172 78aff1a 13d4172 78aff1a 13d4172 78aff1a 13d4172 78aff1a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 | ---
license: mit
base_model: unsloth/Qwen2.5-3B-Instruct
library_name: peft
tags:
- qwen2.5
- unsloth
- lora
- peft
- interview
- software-engineering
- text-generation
language:
- en
pipeline_tag: text-generation
datasets:
- shimogerald/interview-coach-dataset
---
# Interview Coach LoRA (Qwen2.5-3B-Instruct)
LoRA adapter fine-tuned for software-engineering interview Q&A coaching.
## Model Details
- **Base model:** `unsloth/Qwen2.5-3B-Instruct`
- **Method:** QLoRA (4-bit) + LoRA via Unsloth
- **LoRA:** `r=16`, `lora_alpha=16`, `lora_dropout=0`
- **Target modules:** `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj`
- **Context length:** 2048
- **Language:** English
## Training Data
Fine-tuned on [`shimogerald/interview-coach-dataset`](https://huggingface.co/datasets/shimogerald/interview-coach-dataset) (chat `messages` format, ~90/10 train/val).
## Intended Use
Practice / coaching-style answers to technical interview questions (APIs, systems, coding concepts, behavioral, etc.).
## Limitations
- Synthetic training data may contain errors
- Not a substitute for real interview feedback
- May hallucinate technical details
- English only
## How to Use
```python
from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
model_name="shimogerald/lora_interview_coach",
max_seq_length=2048,
load_in_4bit=True,
)
FastLanguageModel.for_inference(model)
messages = [{"role": "user", "content": "What is the difference between PUT and PATCH?"}]
prompt = tokenizer.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=256, do_sample=False)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
```
If loading the adapter separately fails, load the base model then attach this repo with PEFT `PeftModel.from_pretrained`.
## Training Setup (summary)
- Optimizer: AdamW
- LR schedule: cosine with warmup
- Epochs: 3
- Framework: Unsloth + Accelerate + Transformers
This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|