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": 75275.02802240914,
-      "required_pages_per_second_upper_estimate": 0.0065246486461681226
+      "required_token_positions_per_second": 75583.9601460061,
+      "required_pages_per_second_upper_estimate": 0.006551426099660198
     },
     {
       "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": 9409.378502801143,
-      "required_pages_per_second_upper_estimate": 0.003753541767512822
+      "required_token_positions_per_second": 9447.995018250762,
+      "required_pages_per_second_upper_estimate": 0.003768946472894033
     },
     {
       "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": 18818.757005602285,
-      "required_pages_per_second_upper_estimate": 0.004353388234434634
+      "required_token_positions_per_second": 18895.990036501524,
+      "required_pages_per_second_upper_estimate": 0.0043712547368782535
     },
     {
       "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": 28228.13550840343,
-      "required_pages_per_second_upper_estimate": 0.0017189453508806496
+      "required_token_positions_per_second": 28343.985054752284,
+      "required_pages_per_second_upper_estimate": 0.0017259999804376995
     },
     {
       "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": 28228.13550840343,
-      "required_pages_per_second_upper_estimate": 0.0021730800119171964
+      "required_token_positions_per_second": 28343.985054752284,
+      "required_pages_per_second_upper_estimate": 0.002181998430687201
     },
     {
       "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": 28228.13550840343,
-      "required_pages_per_second_upper_estimate": 0.3504989695221256
+      "required_token_positions_per_second": 28343.985054752284,
+      "required_pages_per_second_upper_estimate": 0.35193743316428827
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08514133327339835,
-    "producer_p95_seconds": 0.20805758099959348,
-    "gpu_update_seconds": 0.9794483235270454,
-    "producer_batches_per_second": 11.74517665572476,
-    "gpu_batches_per_second": 1.0209829104601935,
-    "producer_headroom": 11.503793584978606,
+    "producer_batch_seconds": 0.05698920389842499,
+    "producer_p95_seconds": 0.19482784299907507,
+    "gpu_update_seconds": 0.9754450528601449,
+    "producer_batches_per_second": 17.547183178455263,
+    "gpu_batches_per_second": 1.0251730705567232,
+    "producer_headroom": 17.11631302305494,
     "sustainable": true,
-    "buffer_seconds": 60.72579605867681,
+    "buffer_seconds": 60.477593277328985,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430306.79937164066,
+    "generation_bytes_per_second": 430760.7406943461,
     "upload_bytes_per_second": 5227872.862980623,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.33382211553493,
+    "estimated_full_shards_per_hour": 92.43122735617435,
     "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": [

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -105,3 +105,4 @@
 {"step": 165500, "loss": 0.6245028488337994, "lr": 0.0003, "grad_norm": 0.15296132018136824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24221713492, "tokens_per_second": 149292.2538293545, "data_wait_s": 1.1387999620637856e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 165600, "loss": 0.6159810349345207, "lr": 0.0003, "grad_norm": 0.13333768656349962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24236439735, "tokens_per_second": 150376.87033894085, "data_wait_s": 1.4901999747962691e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 165700, "loss": 0.6078481357544661, "lr": 0.0003, "grad_norm": 0.13609650753847735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24251122613, "tokens_per_second": 150290.87363514613, "data_wait_s": 1.2860000424552709e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 165800, "loss": 0.6403736751526594, "lr": 0.0003, "grad_norm": 0.14658243310423927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24265691388, "tokens_per_second": 148917.27265474034, "data_wait_s": 9.196999599225819e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01beb56729fe8cb92e8094373bd8550ced2578a504d5401b25e97cb0685e894b
-size 228357
+oid sha256:6a4409ece63a469b7b07f87b92e27c191beaedeae87a9d140ff1648545a52e93
+size 230475