File size: 3,444 Bytes
3562ec0 d741fc9 3562ec0 767eeab 3562ec0 767eeab 4d41496 |
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 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 |
---
base_model: unsloth/qwen3-14b-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen3
- trl
license: apache-2.0
language:
- en
datasets:
- mattwesney/CoT_Heartbreak_and_Breakups
pipeline_tag: text-generation
library_name: peft
---
# Model Description
- **Developed by:** khazarai
- **License:** apache-2.0
- **Finetuned from model :** unsloth/qwen3-14b-unsloth-bnb-4bit
This qwen3 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
This model is a QLoRA fine-tuned version of unsloth/qwen3-14b-unsloth-bnb-4bit, originally based on the Qwen3-14B architecture developed by the Qwen Team.
The model has been fine-tuned on the Chain of Thought – Heartbreak & Breakups Dataset (MIT Licensed), consisting of 9.8k high-quality Q&A pairs focused on emotional processing, coping strategies, and relationship dynamics following breakups.
The goal of this fine-tuning is to enhance:
- Emotional reasoning capability
- Structured chain-of-thought generation
- Empathetic and psychologically grounded responses
- Relationship pattern analysis
- Identity reconstruction & self-esteem rebuilding guidance
# 🧠 Base Model
- Base architecture: Qwen3-14B
- Variant: unsloth/qwen3-14b-unsloth-bnb-4bit
- Quantization: 4-bit (bitsandbytes)
- Fine-tuning method: QLoRA
- Adapter type: LoRA
- Training precision: 4-bit base + 16-bit adapters
# 🎯 Intended Use
This model is intended for:
- Mental health–adjacent AI assistants
- Relationship guidance systems
- Emotional reasoning research
- Chain-of-thought alignment experiments
- NLP research on structured reasoning in affective domains
The model aims to produce:
- Step-by-step reasoning
- Balanced perspectives
- Reduced reactive or extreme advice
⚠️ Limitations
- Not a substitute for licensed therapy
- May generate plausible but non-clinically validated advice
- Trained on synthetic / curated emotional scenarios
- Chain-of-thought exposure may increase verbosity
- Emotional nuance outside breakup domain may be limited
This model should not be used for crisis intervention or high-risk mental health scenarios.
# How to get started with Model
``` Python
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
tokenizer = AutoTokenizer.from_pretrained("unsloth/qwen3-14b-unsloth-bnb-4bit")
base_model = AutoModelForCausalLM.from_pretrained(
"unsloth/qwen3-14b-unsloth-bnb-4bit",
device_map={"": 0}
)
model = PeftModel.from_pretrained(base_model,"khazarai/Med-R1-14B")
question = """
How can someone work through and move past deeply painful memories associated with trauma, understanding that "moving past" doesn't mean forgetting but rather integrating the experience in a healthy way?
"""
messages = [
{"role" : "user", "content" : question}
]
text = tokenizer.apply_chat_template(
messages,
tokenize = False,
add_generation_prompt = True,
enable_thinking = True,
)
from transformers import TextStreamer
_ = model.generate(
**tokenizer(text, return_tensors = "pt").to("cuda"),
max_new_tokens = 2048,
temperature = 0.6,
top_p = 0.95,
top_k = 20,
streamer = TextStreamer(tokenizer, skip_prompt = True),
)
```
# 🧪 Future Work
- Domain expansion to broader emotional intelligence tasks
- Controlled reasoning output (hidden CoT vs visible CoT)
- Evaluation via human annotation
- Cross-cultural emotional adaptation
|