CoolFace
Modelpublic

leonard-milo/Qwen3.5-0.8B-SFT-AutoConv-InstagramChat

sourceHugging Faceupdated 6mo agoView on Hugging Face
1likes12downloads
Model Card

Qwen3.5-0.8B-SFT-AutoConv-InstagramChat

<!-- Provide a quick summary of what the model is/does. -->

A personalized conversational AI fine-tuned on Instagram chat history using SFT. Designed to generate style-consistent, multi-turn responses that mimic the user's conversational behavior.

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

  • —Developed by: leonard-milo
  • —Funded by: leonard-milo <!-- - Shared by [optional]: [More Information Needed] -->
  • —Model type: Causal language model, conversational AI <!-- - Language(s) (NLP): [More Information Needed] -->
  • —License: [More Information Needed]
  • —Finetuned from model: Qwen-3.5-0.8B

Model Sources

<!-- Provide the basic links for the model. -->

  • —Repository: https://github.com/leonard-cs/Conversation-SFT/tree/main <!-- - Paper [optional]: [More Information Needed] -->
  • —Demo: https://huggingface.co/spaces/leonard-milo/LeonardGPT-chat

Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

  • —Generating personalized chat responses
  • —Mimicking a specific conversational style for research or personal chatbot use
  • —Multi-turn conversation modeling

Downstream Use

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->

  • —Integration into chatbots or digital assistants
  • —Experimental research in style-consistent dialogue generation

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->

  • —This model should not be used for public-facing automated messaging on social media without consent.
  • —Not suitable for factual question answering, safety-critical tasks, or general-purpose LLM replacement.

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

  • —Trained on a personal chat dataset, so output reflects only the user’s style.
  • —Limited generalization outside the context of personal messaging.
  • —May reproduce idiosyncrasies, slang, or informal expressions.
  • —Risk of generating biased or offensive content inherent in informal personal chats.

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

python
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer

MODEL_NAME = "leonard-milo/Qwen3.5-0.8B-SFT-AutoConv-InstagramChat"

tok = AutoTokenizer.from_pretrained(MODEL_NAME, device_map="auto")
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, device_map="auto")

messages = [
    {"role": "user", "content": "User message"},
]

tokenized_chat = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, return_tensors="pt")

inputs = tok(tokenized_chat, return_tensors="pt")
streamer = TextStreamer(tok)
_ = model.generate(**inputs, streamer=streamer)

Training Details

Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

  • —Personal Instagram chat history exported via “Download Your Data”
  • —Preprocessed to remove system messages, attachments, and irrelevant content
  • —Formatted for multi-turn dialogue with clear role separation (Friend: / Me:)
  • —Dataset includes ~32k cleaned message turns

Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

Preprocessing
  • —Removed system messages, attachments, reactions.
  • —Merged consecutive messages of mine (many to one)
  • —Constructed conversation samples using a sliding window of 10 messages per conversation to preserve multi-turn context.
  • —Converted conversations into a structured chat format:
python
  {
    "messages": [
      {"role": "user", "content": "..."},
      {"role": "assistant", "content": "..."}
    ]
  }
  • —Mapped roles:
  • —user → other participant (friend)
  • —assistant → target user (model persona)
Training Hyperparameters
  • —Learning rate: 1e-5
  • —Batch size: 4
  • —epoch: 1

Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

<!-- Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). -->

  • —Hardware Type: Nvidia A30 (24GB)
  • —Hours used: ~2 hours <!-- - Cloud Provider: [More Information Needed] --> <!-- - Compute Region: [More Information Needed] -->
  • —Carbon Emitted: 0.06 kg Estimated using Machine Learning Impact calculator

Technical Specifications

Compute Infrastructure

<!-- [More Information Needed] -->

Hardware

Nvidia A30 (24GB)

Model Card Authors

leonard-milo

Model Card Contact

leonard-milo