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": 76520.19168991936,
-      "required_pages_per_second_upper_estimate": 0.005293844412415138
+      "required_token_positions_per_second": 76991.5144770024,
+      "required_pages_per_second_upper_estimate": 0.005326451616445081
     },
     {
       "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": 9565.02396123992,
-      "required_pages_per_second_upper_estimate": 0.004004675792157604
+      "required_token_positions_per_second": 9623.9393096253,
+      "required_pages_per_second_upper_estimate": 0.004029342418234188
     },
     {
       "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": 19130.04792247984,
-      "required_pages_per_second_upper_estimate": 0.003984284488963131
+      "required_token_positions_per_second": 19247.8786192506,
+      "required_pages_per_second_upper_estimate": 0.004008825515696042
     },
     {
       "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": 28695.07188371976,
-      "required_pages_per_second_upper_estimate": 0.0017202257445911946
+      "required_token_positions_per_second": 28871.817928875906,
+      "required_pages_per_second_upper_estimate": 0.0017308214001227208
     },
     {
       "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": 28695.07188371976,
-      "required_pages_per_second_upper_estimate": 0.002230915622676943
+      "required_token_positions_per_second": 28871.817928875906,
+      "required_pages_per_second_upper_estimate": 0.002244656850264141
     },
     {
       "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": 28695.07188371976,
-      "required_pages_per_second_upper_estimate": 0.36806957175664445
+      "required_token_positions_per_second": 28871.817928875906,
+      "required_pages_per_second_upper_estimate": 0.37033668024878985
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1374304687344079,
-    "producer_p95_seconds": 0.21704952200161642,
-    "gpu_update_seconds": 0.9635103934235544,
-    "producer_batches_per_second": 7.276406820183057,
-    "gpu_batches_per_second": 1.03787152357204,
-    "producer_headroom": 7.010893598024412,
+    "producer_batch_seconds": 0.4001747431562137,
+    "producer_p95_seconds": 0.23397733499950846,
+    "gpu_update_seconds": 0.9576120238811872,
+    "producer_batches_per_second": 2.4989083321773666,
+    "gpu_batches_per_second": 1.0442642480062176,
+    "producer_headroom": 2.39298466546993,
     "sustainable": true,
-    "buffer_seconds": 60.701154785683926,
+    "buffer_seconds": 60.32955750451479,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 241864.6555060695,
-    "upload_bytes_per_second": 5069981.321792331,
+    "generation_bytes_per_second": 239983.06571415733,
+    "upload_bytes_per_second": 5120941.759695578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.898524750581394,
+    "estimated_full_shards_per_hour": 51.49477938240566,
     "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": 1.0705139929996221,
-      "bytes": 6406862,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 3.478892785999051,
       "bytes": 11471341,
@@ -562,6 +556,12 @@
       "bytes": 7439707,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0849376400001347,
+      "bytes": 10733196,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 576200, "loss": 0.616330087184906, "lr": 0.0003, "grad_norm": 0.16241952339423624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84338476900, "tokens_per_second": 152404.30469681544, "data_wait_s": 1.3199001841712743e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 576300, "loss": 0.5840073227882385, "lr": 0.0003, "grad_norm": 0.15837195661544468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84353045730, "tokens_per_second": 150939.17779391387, "data_wait_s": 8.772000001044944e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 576400, "loss": 0.6187487840652466, "lr": 0.0003, "grad_norm": 0.16270405619755987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84367598981, "tokens_per_second": 151528.78434761916, "data_wait_s": 9.620001947041601e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 576500, "loss": 0.6035013794898987, "lr": 0.0003, "grad_norm": 0.16464012626293767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84382130335, "tokens_per_second": 151303.1980639094, "data_wait_s": 1.166500078397803e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e3c13ea2ffc111361d3a6bcfa1fdfb23a72ccb99c05a0e0633ad89409b50466
-size 412071
+oid sha256:f91f0cdefce28a98fbd30e069a81fe28193992c6ca7acc99a59bb1eb183560c8
+size 414189