From d698a32f0f3fb2588f912071de8a80f8ef39f67a Mon Sep 17 00:00:00 2001
From: Christian Medina <37550954+cmedinasoriano@users.noreply.github.com>
Date: Thu, 2 Jul 2026 23:04:52 -0400
Subject: [PATCH] feat: skip already processed shards (resume support)

---
 quantize_streaming.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/quantize_streaming.py b/quantize_streaming.py
index aaf4882..c7ebf62 100644
--- a/quantize_streaming.py
+++ b/quantize_streaming.py
@@ -36,6 +36,12 @@ def streaming_quantize(model_path: str, output_path: str):
     print(f"Found {len(shards)} shards\n")
 
     for idx, shard_file in enumerate(shards):
+        # Skip already processed shards
+        shard_name = f"model-{idx:05d}-of-{len(shards):05d}.safetensors"
+        if (output_path / shard_name).exists():
+            print(f"[{idx+1}/{len(shards)}] {Path(shard_file).name} (already saved, skipping)")
+            continue
+        
         print(f"[{idx+1}/{len(shards)}] {Path(shard_file).name}")
 
         state_dict = load_file(shard_file, device="cuda:0")