kshitijthakkar/moe-141m-89m-8x2-10L-small-250m-8exp

small_250m_8exp

Mobile-optimized MoE model configuration from architecture search.

Metrics

• Best Training Loss: 3.8722
• Best Eval Loss: 6.1125
• Total Parameters: 141.5M
• Active Parameters: 89.9M
• Steps Completed: 2000

Usage

# Load the model
from safetensors.torch import load_file
from architecture.model import Qwen3Model  # Your custom model class

# Load config
import json
with open("config.json") as f:
    config = json.load(f)

# Initialize model
model = Qwen3Model(config)

# Load weights
state_dict = load_file("model.safetensors")
model.load_state_dict(state_dict)

# Load tokenizer
from transformers import AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("kshitijthakkar/moe-141m-89m-8x2-10L-small-250m-8exp")

Model Configuration

{
  "vocab_size": 151936,
  "emb_dim": 448,
  "n_heads": 7,
  "n_layers": 10,
  "n_kv_groups": 1,
  "num_experts": 8,
  "num_experts_per_tok": 2,
  "moe_hidden_dim": 640,
  "head_dim": 64,
  "max_position_embeddings": 4096,
  "rope_base": 1000000.0,
  "qk_norm": true
}

Training Configuration

{
  "model_config": {
    "vocab_size": 151936,
    "emb_dim": 448,
    "n_heads": 7,
    "n_layers": 10,
    "n_kv_groups": 1,
    "num_experts": 8,
    "num_experts_per_tok": 2,
    "moe_hidden_dim": 640,
    "head_dim": 64,
    "max_position_embeddings": 4096,
    "rope_base": 1000000.0,
    "qk_norm": true
  },
  "learning_rate": 0.0001,
  "batch_size": 4,
  "context_length": 1024,
  "warmup_ratio": 0.1,
  "warmup_steps": null,
  "weight_decay": 0.1,
  "gradient_clip": 1.0,
  "gradient_accumulation_steps": 1,
  "scheduler_type": "cosine",
  "wsd_decay_ratio": 0.1,
  "max_steps": 2000,
  "eval_steps": 500,
  "eval_batches": 20,
  "log_steps": 100,
  "early_stopping": true,
  "early_stopping_patience": 500,
  "early_stopping_min_delta": 0.01,
  "early_stopping_min_steps": 200,
  "track_expert_balance": true,
  "expert_balance_log_steps": 100,
  "use_wandb": true,
  "wandb_project": "moe-architecture-search",
  "wandb_entity": null,
  "wandb_tags": [
    "small_250m_8exp",
    "architecture-search"
  ],
  "train_data_path": null,
  "val_data_path": null,
  "output_dir": null,
  "experiment_name": "small_250m_8exp",
  "device": "cuda",
  "dtype": "bfloat16",
  "gradient_checkpointing": true,
  "architecture_name": "small_250m_8exp",
  "mobile_estimate": {
    "tok_per_sec_fp16": 50.26588524585013,
    "tok_per_sec_q8": 83.77647540975022,
    "tok_per_sec_q4": 117.2870655736503,
    "ttft_ms_fp16": 51.45923490909091,
    "ttft_ms_q8": 34.306156606060604,
    "ttft_ms_q4": 27.444925284848484,
    "memory_mb_fp16": 327.3206787109375,
    "memory_mb_q8": 202.15606079101565,
    "memory_mb_q4": 132.82589111328124,
    "total_params": 141512896,
    "active_params": 89903296,
    "meets_ttft_target": true,
    "meets_throughput_target": true,
    "meets_memory_target": true
  }
}