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": 76953.14893354457,
-      "required_pages_per_second_upper_estimate": 0.005323797399128342
+      "required_token_positions_per_second": 76891.06660887785,
+      "required_pages_per_second_upper_estimate": 0.005319502399857059
     },
     {
       "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": 9619.14361669307,
-      "required_pages_per_second_upper_estimate": 0.0040273345617489196
+      "required_token_positions_per_second": 9611.383326109732,
+      "required_pages_per_second_upper_estimate": 0.00402408549013497
     },
     {
       "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": 19238.28723338614,
-      "required_pages_per_second_upper_estimate": 0.00400682788296233
+      "required_token_positions_per_second": 19222.766652219463,
+      "required_pages_per_second_upper_estimate": 0.004003595355210561
     },
     {
       "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": 28857.430850079214,
-      "required_pages_per_second_upper_estimate": 0.0017299589167166566
+      "required_token_positions_per_second": 28834.149978329195,
+      "required_pages_per_second_upper_estimate": 0.0017285632640030757
     },
     {
       "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": 28857.430850079214,
-      "required_pages_per_second_upper_estimate": 0.0022435383181697703
+      "required_token_positions_per_second": 28834.149978329195,
+      "required_pages_per_second_upper_estimate": 0.0022417283327929387
     },
     {
       "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": 28857.430850079214,
-      "required_pages_per_second_upper_estimate": 0.37015213824962756
+      "required_token_positions_per_second": 28834.149978329195,
+      "required_pages_per_second_upper_estimate": 0.36985351622387086
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05566452411726175,
-    "producer_p95_seconds": 0.1858664809988113,
-    "gpu_update_seconds": 0.9580894482130972,
-    "producer_batches_per_second": 17.964763300471596,
-    "gpu_batches_per_second": 1.0437438820196474,
-    "producer_headroom": 17.21185015782773,
+    "producer_batch_seconds": 0.11114019264847741,
+    "producer_p95_seconds": 0.21265031099937914,
+    "gpu_update_seconds": 0.958863015583235,
+    "producer_batches_per_second": 8.997645011853411,
+    "gpu_batches_per_second": 1.0429018366004483,
+    "producer_headroom": 8.627509029213215,
     "sustainable": true,
-    "buffer_seconds": 60.359635237425124,
+    "buffer_seconds": 60.4083699817438,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279098.7733474641,
-    "upload_bytes_per_second": 3445301.604025005,
+    "generation_bytes_per_second": 305453.98348400433,
+    "upload_bytes_per_second": 3396063.340060806,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.88809967344229,
+    "estimated_full_shards_per_hour": 65.54331425085161,
     "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": 1.1770538490000035,
-      "bytes": 8003790,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2935991909998847,
       "bytes": 11779732,
@@ -562,6 +556,12 @@
       "bytes": 7719194,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9053547809999145,
+      "bytes": 813842,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 697500, "loss": 0.5619334578514099, "lr": 8e-05, "grad_norm": 0.20341879759776943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102095144107, "tokens_per_second": 153889.78511646157, "data_wait_s": 1.642700044612866e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 697600, "loss": 0.5490195751190186, "lr": 8e-05, "grad_norm": 0.20628541522345373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102109910778, "tokens_per_second": 154338.96060391763, "data_wait_s": 1.7364000086672604e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 697700, "loss": 0.6124753355979919, "lr": 8e-05, "grad_norm": 0.2217833666166227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102124586313, "tokens_per_second": 153359.24384102013, "data_wait_s": 6.705000487272628e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 697800, "loss": 0.5559768676757812, "lr": 8e-05, "grad_norm": 0.19608597554671559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102139096706, "tokens_per_second": 152363.87423358994, "data_wait_s": 1.2340000466792844e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b0f01a35fd21e74c15a8c17b059ae127aaa5676783faffaf08a1290dc0fdb8f
-size 270717
+oid sha256:36a870d3601277e5f1e3f25798af2d088155a1df26350c95f72c889f634fb3e9
+size 272835