harpertoken
/

chat

@@ -1,108 +1,42 @@
 ---
 license: mit
-language:
-- en
-base_model:
-- gpt2
 tags:
-- text-generation-inference
-- conversational-ai
 - gpt2
-metrics:
-- perplexity
-- bleu
-- f1
-library_name: transformers
 ---
-> **Renamed:** previously `harpertoken/harpertokenConvFT`. The old ID still redirects.
-# HarpertokenConvFT
-## Model Details
-- **Model Name**: HarpertokenConvFT
-- **Base Model**: gpt2
-- **Model Type**: GPT-2-based conversational AI model
-- **Max Sequence Length**: 1024 tokens
-## Intended Use
-Generates human-like responses for chatbots, virtual assistants, and dialogue systems.
-## Training Data
-The model was fine-tuned on the DailyDialog dataset, featuring:
-- **Training Examples**: 11,118
-- **Validation Examples**: 1,000
-- **Test Examples**: 1,000
-## Dataset Characteristics
-- **Description**: A high-quality, multi-turn dialogue dataset covering everyday topics.
-- **Features**: Includes dialogues, communication acts, and emotion annotations.
-- **Citation**:
-  ```
-  @InProceedings{li2017dailydialog,
-      author = {Li, Yanran and Su, Hui and Shen, Xiaoyu and Li, Wenjie and Cao, Ziqiang and Niu, Shuzi},
-      title = {DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset},
-      booktitle = {Proceedings of The 8th International Joint Conference on Natural Language Processing (IJCNLP 2017)},
-      year = {2017}
-  }
-  ```
-## Training Configuration
-- **Learning Rate**: 2e-5
-- **Batch Size**: 8
-- **Number of Epochs**: 3
-- **Weight Decay**: 0.01
-## Ethical Considerations
-Inherited from the GPT-2 base model and the DailyDialog dataset, this model may reflect biases or limitations present in its training data. Caution is advised when using it in sensitive contexts, as it could produce biased or inappropriate responses.
-## How to Use
-### Using the Model Directly
 ```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-# Load model and tokenizer
-model = AutoModelForCausalLM.from_pretrained("harpertoken/harpertokenConvFT")
-tokenizer = AutoTokenizer.from_pretrained("harpertoken/harpertokenConvFT")
-# Prepare input
-input_text = "Hello, how are you?"
-inputs = tokenizer(input_text, return_tensors="pt")
-# Generate response
-outputs = model.generate(**inputs)
-response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-print(response)
 ```
-### Using the Terminal
-Run the provided script to generate responses:
-```bash
-python3 generate_response.py --input "Hello, how are you?"
-```
-### Using the API
-**Check API Status:**
-```bash
-curl http://localhost:8000/status
-```
-**Generate a Response:**
-```bash
-curl -X POST http://localhost:8000/chat -H "Content-Type: application/json" -d '{"input_text": "Hello, how are you?"}'
-```
-### Using FastAPI Documentation
-Interact with the API via the browser at:
-[http://localhost:8000/docs#/default/generate_response_chat_post](http://localhost:8000/docs#/default/generate_response_chat_post)
-## Related Models
-- **harpertokenConvAI**: [https://huggingface.co/harpertoken/harpertokenConvAI](https://huggingface.co/harpertoken/harpertokenConvAI) - DistilBERT-based model for question answering. Note: This is not the base model for harpertokenConvFT due to incompatible architectures (DistilBERT vs GPT-2).
-- **Base Model**: This model is fine-tuned from GPT-2 ([openai/gpt2](https://huggingface.co/gpt2)).
-## Model Differences
-harpertokenConvFT is a GPT-2 model for conversational AI, while harpertokenConvAI is a DistilBERT model for question answering. They have different architectures, tokenizers, and parameters, making fine-tuning between them impossible. For truthful info, refer to the config.json files.

 ---
 license: mit
+language: en
+library_name: transformers
+pipeline_tag: text-generation
+base_model: openai-community/gpt2
+datasets:
+- daily_dialog
 tags:
 - gpt2
+- text-generation
+- dialogue
 ---
+> Previously published as `harpertoken/harpertokenConvFT`. The old identifier still redirects here.
+# chat
+`openai-community/gpt2` further fine-tuned on DailyDialog, a corpus of short multi-turn English conversations about everyday topics. The base model is the 124M-parameter GPT-2 — `config.json` records `_name_or_path: gpt2`, twelve layers, 768 embedding dimensions. The previous version of this card declared its base model as `gpt2`, an identifier that has since moved; `openai-community/gpt2` is the current one.
+This is a continuation, not a chat model. Nothing here establishes turn-taking, instruction following or any conversational protocol: there is no chat template in `tokenizer_config.json` and no special tokens beyond the end-of-text marker. It produces fluent English that resembles dialogue because that is what it was trained to imitate, which is a weaker property than it sounds.
+The card this replaces documented a `generate_response.py` script and a FastAPI service on `localhost:8000` with `/status`, `/chat` and interactive docs endpoints. None of that is in the repository — it contains weights and tokenizer files only — so those sections have been removed. The declared perplexity, BLEU and F1 metrics likewise had no values behind them; no evaluation is recorded for this checkpoint.
+## Usage
 ```python
+from transformers import pipeline
+generator = pipeline("text-generation", model="harpertoken/chat")
+print(generator("Hello, how are you?", max_new_tokens=60)[0]["generated_text"])
 ```
+A `pad_token` is not configured, so batching prompts together will warn. Set `tokenizer.pad_token = tokenizer.eos_token` if you need it.
+## Limitations
+DailyDialog is scripted, crowd-sourced, and narrow — 13k conversations of polite small talk, annotated for emotion and communication acts. A model trained on it will handle greetings and farewells far better than disagreement, and will reproduce the register of that corpus rather than anything wider. Because GPT-2 is a 124M-parameter base and the fine-tuning is small, output is frequently fluent and wrong. The biases of both GPT-2 and DailyDialog carry through unchanged.
+## Attribution
+GPT-2 follows Radford et al. (2018). DailyDialog is described in Li et al., *DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset*, IJCNLP 2017.