ANLP Assignment 2 (part1): moe_4e1a

Decoder-only transformer with Mixture-of-Experts feed-forward, 4 experts, top-1 routing.

Trained from scratch for IIIT-H Advanced NLP, Monsoon 2026. Architecture, training code and evaluation live in the assignment repository. This repository holds the checkpoint and the tokenizer it was trained with, which every model in the assignment shares.

Architecture

Setting Value
d_model 512
n_layers 8
n_heads 8
n_kv_heads 8
n_ctx 256
vocab_size 32000
ffn moe_4e1a
d_ff 512
n_routed_experts 4
n_shared_experts 0
top_k 1
norm rmsnorm

Parameters

Count Value
Total 41,607,680
Active per token 29,000,192
Feed-forward (total / active) 16,809,984 / 4,202,496

Test set results

Metric Value
Perplexity (both directions) 10.59
BLEU vi->en 33.66
BLEU ja->en 23.13
BLEU mean 28.39

Training

  • Scored target tokens: 33,393,292
  • Best validation loss: 2.1011
  • Best validation perplexity: 8.18

Loading

import torch
payload = torch.load("model.pt", map_location="cpu", weights_only=False)
state_dict, config = payload["model"], payload["config"]

from tokenizers import Tokenizer
tokenizer = Tokenizer.from_file("tokenizer.json")
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support