Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77186.35172105877,
-      "required_pages_per_second_upper_estimate": 0.005339930909073591
+      "required_token_positions_per_second": 76670.38344338864,
+      "required_pages_per_second_upper_estimate": 0.005304235026413024
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9648.293965132347,
-      "required_pages_per_second_upper_estimate": 0.004039539203911948
+      "required_token_positions_per_second": 9583.79793042358,
+      "required_pages_per_second_upper_estimate": 0.004012536061009745
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19296.587930264694,
-      "required_pages_per_second_upper_estimate": 0.004018970380629364
+      "required_token_positions_per_second": 19167.59586084716,
+      "required_pages_per_second_upper_estimate": 0.0039921047343193196
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28944.881895397044,
-      "required_pages_per_second_upper_estimate": 0.0017352014733534459
+      "required_token_positions_per_second": 28751.39379127074,
+      "required_pages_per_second_upper_estimate": 0.0017236021569503024
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28944.881895397044,
-      "required_pages_per_second_upper_estimate": 0.0022503372522867347
+      "required_token_positions_per_second": 28751.39379127074,
+      "required_pages_per_second_upper_estimate": 0.0022352944032551416
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28944.881895397044,
-      "required_pages_per_second_upper_estimate": 0.3712738663613479
+      "required_token_positions_per_second": 28751.39379127074,
+      "required_pages_per_second_upper_estimate": 0.3687920087129557
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.23294641195305132,
-    "producer_p95_seconds": 0.19772154899965244,
-    "gpu_update_seconds": 0.9551947767455469,
-    "producer_batches_per_second": 4.292832809124971,
-    "gpu_batches_per_second": 1.0469068972582842,
-    "producer_headroom": 4.100491476718085,
+    "producer_batch_seconds": 0.08136120178907191,
+    "producer_p95_seconds": 0.17757504600012908,
+    "gpu_update_seconds": 0.9616229460289419,
+    "producer_batches_per_second": 12.290870562513197,
+    "gpu_batches_per_second": 1.0399086296032531,
+    "producer_headroom": 11.81918315958434,
     "sustainable": true,
-    "buffer_seconds": 60.17727093496945,
+    "buffer_seconds": 60.582245599823345,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429509.5020322878,
-    "upload_bytes_per_second": 2376254.5933714784,
+    "generation_bytes_per_second": 412533.11110252567,
+    "upload_bytes_per_second": 2362602.022411542,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.16274066664197,
+    "estimated_full_shards_per_hour": 88.52000236327007,
     "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": [
@@ -328,6 +328,12 @@
       "bytes": 9407988,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.460241586999473,
+      "bytes": 2517815,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -62,3 +62,4 @@
 {"step": 651200, "loss": 0.5152514576911926, "lr": 0.0001, "grad_norm": 0.19390245787186547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95318582360, "tokens_per_second": 153970.00772675822, "data_wait_s": 1.3885000043956097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 651300, "loss": 0.5093046426773071, "lr": 0.0001, "grad_norm": 0.1805517497239884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95333195261, "tokens_per_second": 152386.52056084806, "data_wait_s": 1.4483000086329412e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 651400, "loss": 0.5146275162696838, "lr": 0.0001, "grad_norm": 0.17616883334889957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95347576330, "tokens_per_second": 149689.76331282474, "data_wait_s": 1.3550999938161112e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 651500, "loss": 0.5590850114822388, "lr": 0.0001, "grad_norm": 0.19672708800431543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95361996926, "tokens_per_second": 150293.79136367328, "data_wait_s": 1.3192999176681042e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d19ba2803d3f724ce909d019934296dd19fc26e268344bd4bc12875fd5bc570
-size 136500
+oid sha256:58cd5d4b91bd876f2c6d8aacbdb957120302c7ded96d1c10a2cfc853863ad58a
+size 138618