Booper-Big-Chat
Booper-Big-Chat is Booper-Big supervised-fine-tuned on mookiezi/Discord-Dialogues with
assistant-only loss. Its final 25,023,805 assistant tokens
used symmetric per-output-channel INT8 fake-quantization-aware training. This repository keeps the
learned values in broadly compatible BF16; the separate ProCreations/Booper-Big-Chat-INT8
repository stores the weights as real INT8 tensors.
| Property | Value |
|---|---|
| Total / active parameters | 149,602,432 / 50,512,000 |
| Context window | 4,096 |
| Assistant SFT tokens | 100,023,805 |
| QAT scheme | symmetric per-output-channel INT8 weights |
| Final SFT loss | 2.4934 |
| Held-out Discord loss | 2.2578 |
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "ProCreations/Booper-Big-Chat"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto")
messages = [{"role": "user", "content": "hey, what's up?"}]
inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
output = model.generate(inputs, max_new_tokens=80, temperature=0.8, do_sample=True)
print(tokenizer.decode(output[0, inputs.shape[1]:], skip_special_tokens=True))
Validation sample for hey, what's up?: I'm tired, and is it just me or am I okay? How tired?
The Discord corpus is informal internet dialogue. The model can reproduce profanity, bias, personal data patterns, and unreliable claims. It is not suitable for high-stakes use.