---
license: odc-by
language:
- en
library_name: transformers
tags:
- ivmelabs
- causal-lm
- sft
- conversational
- chatmaxxing
pipeline_tag: text-generation
extra_gated_prompt: >-
  This is a checkpoint bucket for an in-progress SFT training run
  ("Chatmaxxing" -- finetuning Ivme-Conversate-U-v1-Base on conversational
  data). It is gated to manual approval, not because the content is
  sensitive, but to keep casual downloads of intermediate/incomplete
  training checkpoints separate from the polished, final public release
  this run is working towards. Requests are generally approved -- just ask.
---

# Ivme-Conversate-Chat-v1 (in progress / checkpoint bucket)

This repository holds intermediate and final checkpoints from an active SFT
finetuning run: taking [Ivme-Conversate-U-v1-Base](https://huggingface.co/IvmeLabs/Ivme-Conversate-U-v1-Base),
a 317M-parameter base model trained on pure distillation data (Cosmopedia v2,
generated by Mixtral-8x7B-Instruct-v0.1), and finetuning it on
[smol-smoltalk](https://huggingface.co/datasets/HuggingFaceTB/smol-smoltalk),
a conversational SFT dataset purpose-built for sub-1B-parameter models (core
component generated by Llama-3.1-405B-Instruct via the Magpie pipeline).

**This repo is gated to manual approval** so that casual traffic doesn't land
on an intermediate, possibly-broken checkpoint mid-run. It is not gated
because of any sensitive content -- access requests are generally approved
quickly.

Checkpoints here may be incomplete, may not follow instructions well yet, and
may be superseded by later commits to this same repo as training continues.
For a stable, finished release, watch the IvmeLabs organization page instead.