Modelpublic

mhla/gpt1900-d34-discovery-rl-v1

sourceHugging Facemitupdated 7mo agoView on Hugging Face

0likes

Model Card

GPT-1900 D34 Discovery RL v1

3.29B parameter GPT-1900 model with discovery RL (v1). Pipeline: base -> reasoning SFT v1 -> discovery RL.

Model Details

—Architecture: Custom GPT with RoPE, QK-norm, ReLU², value embeddings (ResFormer), per-layer residual/skip scalars
—Parameters: 3.29B
—Layers: 34
—Hidden dim: 2176
—Attention heads: 17 (query) / 17 (kv)
—Head dim: 128
—Context length: 2048 tokens
—Vocab size: 32,768 (BPE, GPT-4 style split pattern)
—Training: Discovery RL v1 (104 steps), from reasoning-sft-v1

Checkpoint Contents

model_000104.pt          # Model weights
meta_000104.json         # Training config and metadata
optim_000104_rank*.pt    # Optimizer state shards (if present, for resuming training)
tokenizer/                   # BPE tokenizer (tiktoken format) + token byte counts
nanochat/                    # Source code to load and run the model

Quick Start

python

import torch, json
from nanochat.gpt import GPT, GPTConfig
from nanochat.tokenizer import RustBPETokenizer

tokenizer = RustBPETokenizer.from_directory("tokenizer")

with open("meta_000104.json") as f:
    meta = json.load(f)

config = GPTConfig(**meta["model_config"])

with torch.device("meta"):
    model = GPT(config)
model.to_empty(device="cuda")
model.init_weights()

state_dict = torch.load("model_000104.pt", map_location="cuda")
state_dict = {k.removeprefix("_orig_mod."): v for k, v in state_dict.items()}
model.load_state_dict(state_dict, strict=True, assign=True)
model.eval()

bos = tokenizer.get_bos_token_id()
tokens = tokenizer.encode("It was a dark and stormy night", prepend=bos)
with torch.amp.autocast(device_type="cuda", dtype=torch.bfloat16):
    for token in model.generate(tokens, max_tokens=100, temperature=0.8):
        print(tokenizer.decode([token]), end="", flush=True)

Dependencies

torch>=2.9
tiktoken
rustbpe