From d3e1b0ef1aac9462cf329ad7732010a5448af857 Mon Sep 17 00:00:00 2001
From: Dan Saunders <danjsaund@gmail.com>
Date: Mon, 22 Sep 2025 23:13:45 -0400
Subject: [PATCH] small deepseek script

---
 scripts/benchmarks/build_deepseek_v3_8b.py | 100 +++++++++++++++++++++
 1 file changed, 100 insertions(+)
 create mode 100755 scripts/benchmarks/build_deepseek_v3_8b.py

diff --git a/scripts/benchmarks/build_deepseek_v3_8b.py b/scripts/benchmarks/build_deepseek_v3_8b.py
new file mode 100755
index 000000000..4b0ee063f
--- /dev/null
+++ b/scripts/benchmarks/build_deepseek_v3_8b.py
@@ -0,0 +1,100 @@
+#!/usr/bin/env python3
+"""Instantiate a ~8.3B DeepSeek-V3 MoE model with random weights.
+
+Run this on a GPU-equipped machine (e.g. 1× NVL H100) so the dense
+initialization completes quickly:
+
+    python scripts/benchmarks/build_deepseek_v3_8b.py --output deepseek-v3-8b-moe
+"""
+
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+
+import torch
+from transformers import DeepseekV3Config, DeepseekV3ForCausalLM
+
+DTYPE_MAP = {
+    "float32": torch.float32,
+    "bfloat16": torch.bfloat16,
+    "float16": torch.float16,
+}
+
+
+def build_config() -> DeepseekV3Config:
+    """Return a DeepSeek V3 configuration totaling ~8.3B parameters."""
+
+    return DeepseekV3Config(
+        vocab_size=32_000,
+        hidden_size=3_072,
+        intermediate_size=8_192,
+        moe_intermediate_size=2_560,
+        num_hidden_layers=20,
+        num_attention_heads=24,
+        num_key_value_heads=24,
+        n_routed_experts=18,
+        num_experts_per_tok=4,
+        n_group=6,
+        topk_group=4,
+        kv_lora_rank=192,
+        q_lora_rank=384,
+        max_position_embeddings=2_048,
+        rope_theta=10_000.0,
+        rope_interleave=True,
+        hidden_act="silu",
+        initializer_range=0.02,
+        attention_dropout=0.0,
+        attention_bias=False,
+        n_shared_experts=1,
+        routed_scaling_factor=2.5,
+        norm_topk_prob=True,
+    )
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--output",
+        type=Path,
+        required=True,
+        help="Directory to save the generated model",
+    )
+    parser.add_argument(
+        "--dtype",
+        default="bfloat16",
+        choices=DTYPE_MAP.keys(),
+        help="Storage dtype for the checkpoint",
+    )
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=0,
+        help="Torch RNG seed for reproducibility",
+    )
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+    torch.manual_seed(args.seed)
+
+    output_dir = args.output
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    config = build_config()
+    model = DeepseekV3ForCausalLM(config)
+
+    dtype = DTYPE_MAP[args.dtype]
+    model.to(dtype=dtype)
+
+    param_count = sum(p.numel() for p in model.parameters())
+    print(f"Initialized DeepSeek-V3 MoE with {param_count / 1e9:.3f}B parameters")
+
+    model.save_pretrained(output_dir, safe_serialization=True)
+    config.save_pretrained(output_dir)
+    print(f"Saved model and config to {output_dir.resolve()}")
+
+
+if __name__ == "__main__":
+    main()