Update README and 1 associated file(s)

README.md

@@ -492,6 +492,5 @@ configs:
   - split: train
     path: shard_6/train-*
 ---
-
 # WIP encoder/decoder trial
 Waiting for statistics to finish generating...

runs/shard_2.json

@@ -0,0 +1,79 @@
+{
+ "model_name": "nvidia/Llama-3.3-70B-Instruct-NVFP4",
+ "engine": "vllm",
+ "generation_params": {
+  "temperature": 0.6,
+  "top_p": 0.9,
+  "extra_body": {
+   "chat_template_kwargs": {
+    "enable_thinking": false
+   }
+  }
+ },
+ "ignored_params": {},
+ "prompt_tokens": 3279089,
+ "completion_tokens": 1648480,
+ "failed_requests": 2,
+ "runtime_seconds": 10629.6,
+ "stage": "decode",
+ "decoder_index": 2,
+ "rows_in": 2700,
+ "rows_kept": 2671,
+ "rows_trashed": 29,
+ "dropped": {
+  "empty or too short": 2,
+  "refusal": 2,
+  "filler output": 0,
+  "unfilled placeholder": 25,
+  "task meta-commentary": 0,
+  "echoed instruction": 0,
+  "identical to source": 0
+ },
+ "per_encoder": {
+  "0": {
+   "kept": 298,
+   "trashed": 2,
+   "skipped_empty_encoding": 0
+  },
+  "1": {
+   "kept": 296,
+   "trashed": 4,
+   "skipped_empty_encoding": 0
+  },
+  "2": {
+   "kept": 298,
+   "trashed": 2,
+   "skipped_empty_encoding": 0
+  },
+  "3": {
+   "kept": 297,
+   "trashed": 3,
+   "skipped_empty_encoding": 0
+  },
+  "4": {
+   "kept": 296,
+   "trashed": 4,
+   "skipped_empty_encoding": 0
+  },
+  "5": {
+   "kept": 297,
+   "trashed": 3,
+   "skipped_empty_encoding": 0
+  },
+  "6": {
+   "kept": 294,
+   "trashed": 6,
+   "skipped_empty_encoding": 0
+  },
+  "8": {
+   "kept": 299,
+   "trashed": 1,
+   "skipped_empty_encoding": 0
+  },
+  "9": {
+   "kept": 296,
+   "trashed": 4,
+   "skipped_empty_encoding": 0
+  }
+ }
+}