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": 75210.37284310469,
-      "required_pages_per_second_upper_estimate": 0.006519044499093067
+      "required_token_positions_per_second": 75140.29253136733,
+      "required_pages_per_second_upper_estimate": 0.006512970115288599
     },
     {
       "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": 9401.296605388086,
-      "required_pages_per_second_upper_estimate": 0.0037503177777996194
+      "required_token_positions_per_second": 9392.536566420917,
+      "required_pages_per_second_upper_estimate": 0.0037468232672813614
     },
     {
       "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": 18802.593210776173,
-      "required_pages_per_second_upper_estimate": 0.004349649024974686
+      "required_token_positions_per_second": 18785.073132841833,
+      "required_pages_per_second_upper_estimate": 0.004345596063287417
     },
     {
       "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": 28203.889816164257,
-      "required_pages_per_second_upper_estimate": 0.0017174689154306
+      "required_token_positions_per_second": 28177.609699262754,
+      "required_pages_per_second_upper_estimate": 0.001715868594192416
     },
     {
       "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": 28203.889816164257,
-      "required_pages_per_second_upper_estimate": 0.0021712135114122532
+      "required_token_positions_per_second": 28177.609699262754,
+      "required_pages_per_second_upper_estimate": 0.0021691903952651558
     },
     {
       "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": 28203.889816164257,
-      "required_pages_per_second_upper_estimate": 0.3501979191696271
+      "required_token_positions_per_second": 28177.609699262754,
+      "required_pages_per_second_upper_estimate": 0.3498716080716038
     }
   ],
   "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": 0.4098196449454008,
-    "producer_p95_seconds": 0.3479973030007386,
-    "gpu_update_seconds": 0.9802903138613999,
-    "producer_batches_per_second": 2.4400977657701777,
-    "gpu_batches_per_second": 1.0201059684665892,
-    "producer_headroom": 2.392004204659348,
+    "producer_batch_seconds": 0.10322347237504914,
+    "producer_p95_seconds": 0.3313306839991128,
+    "gpu_update_seconds": 0.9812045909885462,
+    "producer_batches_per_second": 9.687719052568095,
+    "gpu_batches_per_second": 1.0191554434050474,
+    "producer_headroom": 9.505634410587025,
     "sustainable": true,
-    "buffer_seconds": 60.77799945940679,
+    "buffer_seconds": 60.83468464128986,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345746.6315138387,
+    "generation_bytes_per_second": 347324.7302071036,
     "upload_bytes_per_second": 4523427.549510024,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.18917855321284,
+    "estimated_full_shards_per_hour": 74.52780179653007,
     "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

@@ -274,3 +274,4 @@
 {"step": 182400, "loss": 0.6093626208603382, "lr": 0.0003, "grad_norm": 0.13811382871818215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26695548175, "tokens_per_second": 150005.00997656528, "data_wait_s": 1.1318003089400008e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182500, "loss": 0.5690085627138615, "lr": 0.0003, "grad_norm": 0.14368584495889392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26710330108, "tokens_per_second": 151077.2180347098, "data_wait_s": 1.2301999959163368e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 182600, "loss": 0.5898476075381041, "lr": 0.0003, "grad_norm": 0.16368825669799653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26725087444, "tokens_per_second": 150512.07012116656, "data_wait_s": 1.1481999536044896e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 182700, "loss": 0.5470570288598537, "lr": 0.0003, "grad_norm": 0.1333711140535544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26739831087, "tokens_per_second": 151449.40135941628, "data_wait_s": 1.5121000615181401e-05, "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:f291204c235d0ec299e3781cc5baa8bae765edca9d4c4d2b3a8fe5b5e5193f35
-size 588648
+oid sha256:8f7f80d5399ceb1c3682a0d2ef735676dc836c7d25adf2eda0ed53ee3684acb4
+size 590766