Instructions to use lebiraja/claritymentor-lora-failed with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use lebiraja/claritymentor-lora-failed with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("unsloth/qwen2.5-1.5b-instruct-unsloth-bnb-4bit")
model = PeftModel.from_pretrained(base_model, "lebiraja/claritymentor-lora-failed")

Transformers

How to use lebiraja/claritymentor-lora-failed with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="lebiraja/claritymentor-lora-failed")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("lebiraja/claritymentor-lora-failed", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use lebiraja/claritymentor-lora-failed with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "lebiraja/claritymentor-lora-failed"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lebiraja/claritymentor-lora-failed",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/lebiraja/claritymentor-lora-failed

SGLang

How to use lebiraja/claritymentor-lora-failed with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "lebiraja/claritymentor-lora-failed" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lebiraja/claritymentor-lora-failed",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "lebiraja/claritymentor-lora-failed" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "lebiraja/claritymentor-lora-failed",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Studio new

How to use lebiraja/claritymentor-lora-failed with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for lebiraja/claritymentor-lora-failed to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for lebiraja/claritymentor-lora-failed to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for lebiraja/claritymentor-lora-failed to start chatting

Load model with FastModel

pip install unsloth
from unsloth import FastModel
model, tokenizer = FastModel.from_pretrained(
    model_name="lebiraja/claritymentor-lora-failed",
    max_seq_length=2048,
)

Docker Model Runner
How to use lebiraja/claritymentor-lora-failed with Docker Model Runner:
```
docker model run hf.co/lebiraja/claritymentor-lora-failed
```

ClarityMentor LoRA (Failed Experiment) ❌

⚠️ STATUS: FAILED - This model demonstrates what NOT to do when fine-tuning a life mentor LLM.

What This Is

A QLoRA fine-tune of Qwen/Qwen2.5-1.5B-Instruct intended to be a philosophical life mentor called "ClarityMentor".

Why It Failed

The model only asks questions regardless of user input. Even when asked "what is apple" or "hello", it responds with philosophical clarifying questions.

Root Causes:

Overweighted question-asking data: The conversation_starters dataset (2,999 samples) was designed to train Socratic questioning behavior
Imbalanced training mix: Too many datasets emphasized asking clarifying questions
System prompt mismatch: Training data used a question-heavy system prompt, but we wanted direct answers
Overfitting to patterns: Model memorized a narrow set of question templates instead of learning diverse responses

Evidence of Failure:

User: hello
Model: What does success look like to you personally?

User: what is apple
Model: What makes certain decisions feel harder than others for you?

User: give me python code
Model: What specifically draws you to this question right now?

User: this world shall burn to ashes
Model: What does success look like to you personally?

The model ignores user input entirely and outputs memorized question templates.

Training Details

Parameter	Value
Base Model	Qwen/Qwen2.5-1.5B-Instruct
Method	QLoRA (4-bit NF4 quantization)
Hardware	NVIDIA RTX 4050 Laptop GPU (6GB VRAM)
Training Time	~3 hours
Training Samples	27,362
Eval Samples	1,440
Epochs	2
Final Train Loss	0.789
Final Eval Loss	0.68

Dataset Mix (the problem):

Dataset	Samples	Issue
Philosophy QA	11,613	OK - direct Q&A
Counseling	~10,000	Question-heavy responses
Reddit	~10,000	OK - varied
Quotes	4,995	OK - wisdom-based
Conversation Starters	2,999	🔴 Main culprit - trained ONLY to ask questions

LoRA Config:

r: 16
lora_alpha: 32
lora_dropout: 0
target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj
max_seq_length: 512

Training Args:

per_device_train_batch_size: 1
gradient_accumulation_steps: 16
learning_rate: 2e-4
lr_scheduler_type: cosine
warmup_ratio: 0.03
bf16: true
optim: paged_adamw_8bit

Lessons Learned

Audit your training data - The conversation_starters dataset was specifically designed to output clarifying questions, which dominated the learned behavior
Balance response types - Need diverse outputs (questions, answers, advice) not just one pattern
System prompt matters during training - Should match intended inference behavior
Test incrementally - Should have validated on a small sample before full training
Watch for overfitting patterns - Low loss doesn't mean good model behavior

How to Fix (for future attempts)

❌ Remove or significantly reduce conversation_starters dataset
✅ Add more datasets with direct, substantive answers
✅ Include the "no questions" system prompt in training data
✅ Better balance between question-asking and answer-giving examples
✅ Add response diversity validation before training

Usage (not recommended, but here's how)

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    "YOUR_USERNAME/claritymentor-lora-failed",
    max_seq_length=512,
    load_in_4bit=True,
)
FastLanguageModel.for_inference(model)

messages = [
    {"role": "system", "content": "You are ClarityMentor, a philosophical mentor."},
    {"role": "user", "content": "What is the meaning of life?"}
]

inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda")
outputs = model.generate(inputs, max_new_tokens=256, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
# Output: Will ask you questions instead of answering...

Framework Versions

PEFT: 0.18.1
Transformers: 4.57.3
Unsloth: 2026.1.3
PyTorch: 2.9.1+cu128

License

Apache 2.0

This model is shared as a learning resource. Sometimes failures teach more than successes.

Downloads last month: -