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": 75953.79121860611,
-      "required_pages_per_second_upper_estimate": 0.006583482119704884
+      "required_token_positions_per_second": 75525.60146234422,
+      "required_pages_per_second_upper_estimate": 0.006546367716869118
     },
     {
       "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": 9494.223902325764,
-      "required_pages_per_second_upper_estimate": 0.0037873878659349628
+      "required_token_positions_per_second": 9440.700182793027,
+      "required_pages_per_second_upper_estimate": 0.003766036453962433
     },
     {
       "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": 18988.44780465153,
-      "required_pages_per_second_upper_estimate": 0.004392643214338614
+      "required_token_positions_per_second": 18881.400365586054,
+      "required_pages_per_second_upper_estimate": 0.0043678796732813975
     },
     {
       "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": 28482.67170697729,
-      "required_pages_per_second_upper_estimate": 0.0017344452699255726
+      "required_token_positions_per_second": 28322.10054837908,
+      "required_pages_per_second_upper_estimate": 0.0017246673288187026
     },
     {
       "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": 28482.67170697729,
-      "required_pages_per_second_upper_estimate": 0.0021926749130847063
+      "required_token_positions_per_second": 28322.10054837908,
+      "required_pages_per_second_upper_estimate": 0.0021803137008064005
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28482.67170697729,
-      "required_pages_per_second_upper_estimate": 0.35365945723055603
+      "required_token_positions_per_second": 28322.10054837908,
+      "required_pages_per_second_upper_estimate": 0.3516657008378643
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06024325273433817,
-    "producer_p95_seconds": 0.19977702899996075,
-    "gpu_update_seconds": 0.9706954559753317,
-    "producer_batches_per_second": 16.599369300489446,
-    "gpu_batches_per_second": 1.0301892255127783,
-    "producer_headroom": 16.112932352041526,
+    "producer_batch_seconds": 0.10237770528905799,
+    "producer_p95_seconds": 0.22364143199956743,
+    "gpu_update_seconds": 0.9761987799164967,
+    "producer_batches_per_second": 9.767751652340257,
+    "gpu_batches_per_second": 1.0243815302509793,
+    "producer_headroom": 9.535267245541904,
     "sustainable": true,
-    "buffer_seconds": 60.183118270470565,
+    "buffer_seconds": 60.524324354822795,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 211571.3638067165,
-    "upload_bytes_per_second": 4980972.198067309,
+    "generation_bytes_per_second": 207301.21603434748,
+    "upload_bytes_per_second": 5060963.016883097,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 45.398289543639386,
+    "estimated_full_shards_per_hour": 44.482015235641654,
     "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": 1.2415246609998576,
-      "bytes": 5982796,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.1530046430000311,
       "bytes": 2613544,
@@ -562,6 +556,12 @@
       "bytes": 12008396,
       "pages": 45,
       "shards": 4
+    },
+    {
+      "seconds": 1.228597503999481,
+      "bytes": 13077636,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 8800, "loss": 0.7235393300652504, "lr": 0.0003, "grad_norm": 0.17610486994986296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1286956254, "tokens_per_second": 131472.96362443778, "data_wait_s": 1.1267999070696533e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 8900, "loss": 0.7340329103171825, "lr": 0.0003, "grad_norm": 0.1643783651742324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1301647101, "tokens_per_second": 148835.91920559725, "data_wait_s": 1.1284000720479526e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9000, "loss": 0.7084479369223118, "lr": 0.0003, "grad_norm": 0.15896695598011765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1316240288, "tokens_per_second": 149722.04308527088, "data_wait_s": 1.7378000848111697e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 9100, "loss": 0.6799753941595554, "lr": 0.0003, "grad_norm": 0.15308886010203326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1330706081, "tokens_per_second": 146366.16450904598, "data_wait_s": 1.2032998711219989e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a478709e9b7014c917883c41c04efacc04cd18f843fe4675ae23b66ee1e162c9
-size 188369
+oid sha256:2ac233a62e121fc2e432c08166ca871431fc4081e00977e607b477a612f69108
+size 190452