Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74707.1110467094,
-      "required_pages_per_second_upper_estimate": 0.0064754230420868844
+      "required_token_positions_per_second": 74525.46017424166,
+      "required_pages_per_second_upper_estimate": 0.006459678004851308
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9338.388880838675,
-      "required_pages_per_second_upper_estimate": 0.0037252229459225606
+      "required_token_positions_per_second": 9315.682521780207,
+      "required_pages_per_second_upper_estimate": 0.0037161650398038165
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18676.77776167735,
-      "required_pages_per_second_upper_estimate": 0.004320543835101951
+      "required_token_positions_per_second": 18631.365043560414,
+      "required_pages_per_second_upper_estimate": 0.004310038402002139
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28015.166642516022,
-      "required_pages_per_second_upper_estimate": 0.0017059766643093915
+      "required_token_positions_per_second": 27947.047565340617,
+      "required_pages_per_second_upper_estimate": 0.0017018285699025866
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28015.166642516022,
-      "required_pages_per_second_upper_estimate": 0.002156685079085631
+      "required_token_positions_per_second": 27947.047565340617,
+      "required_pages_per_second_upper_estimate": 0.0021514410839589943
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28015.166642516022,
-      "required_pages_per_second_upper_estimate": 0.34785460896874754
+      "required_token_positions_per_second": 27947.047565340617,
+      "required_pages_per_second_upper_estimate": 0.3470087980101148
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2887545934531772,
-    "producer_p95_seconds": 0.2855890040009399,
-    "gpu_update_seconds": 0.9868940046939143,
-    "producer_batches_per_second": 3.463148371221164,
-    "gpu_batches_per_second": 1.0132800434937799,
-    "producer_headroom": 3.417760364923661,
+    "producer_batch_seconds": 0.13161937139071256,
+    "producer_p95_seconds": 0.3195129529995029,
+    "gpu_update_seconds": 0.9892994934566365,
+    "producer_batches_per_second": 7.597665825583504,
+    "gpu_batches_per_second": 1.0108162458528869,
+    "producer_headroom": 7.516366952702558,
     "sustainable": true,
-    "buffer_seconds": 60.200534286328775,
+    "buffer_seconds": 60.347269100854824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540630.5717055602,
-    "upload_bytes_per_second": 3580847.63563265,
+    "generation_bytes_per_second": 535948.6288296074,
+    "upload_bytes_per_second": 3543240.729573944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.00673545241456,
+    "estimated_full_shards_per_hour": 115.00209950128713,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.438195700000506,
-      "bytes": 12388203,
-      "pages": 31,
-      "shards": 4
-    },
     {
       "seconds": 0.9514959919997636,
       "bytes": 336067,
@@ -562,6 +556,12 @@
       "bytes": 6333046,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5801954900016426,
+      "bytes": 8475194,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -226,3 +226,4 @@
 {"step": 217600, "loss": 0.6607991196215153, "lr": 0.0003, "grad_norm": 0.18329649626859376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31849671581, "tokens_per_second": 148777.65195918386, "data_wait_s": 1.2898999557364732e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217700, "loss": 0.5840618945658207, "lr": 0.0003, "grad_norm": 0.14450991405010163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31864315380, "tokens_per_second": 148098.04722320035, "data_wait_s": 1.3136999768903479e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 217800, "loss": 0.5621696505695581, "lr": 0.0003, "grad_norm": 0.1469949327678859, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31878829543, "tokens_per_second": 147603.85871186946, "data_wait_s": 1.3183998817112297e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 217900, "loss": 0.6120729297399521, "lr": 0.0003, "grad_norm": 0.1569615759045584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31893514828, "tokens_per_second": 149033.36150266792, "data_wait_s": 1.341800088994205e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5069d7068b8048c4e1689824cff7e1b81a5b672947aac10b9fc2db4c0495df82
-size 486201
+oid sha256:e1929dc9cacd8c4d19c4a1c5accf4b3d3fbbfb001eb132928cb1b214e5c3ea8e
+size 488319