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": 76933.87353594105,
-      "required_pages_per_second_upper_estimate": 0.005322463882397046
+      "required_token_positions_per_second": 76644.662474337,
+      "required_pages_per_second_upper_estimate": 0.005302455590093161
     },
     {
       "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": 9616.734191992631,
-      "required_pages_per_second_upper_estimate": 0.004026325785941354
+      "required_token_positions_per_second": 9580.582809292126,
+      "required_pages_per_second_upper_estimate": 0.004011189956931369
     },
     {
       "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": 19233.468383985262,
-      "required_pages_per_second_upper_estimate": 0.004005824243713731
+      "required_token_positions_per_second": 19161.16561858425,
+      "required_pages_per_second_upper_estimate": 0.003990765484432848
     },
     {
       "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": 28850.20257597789,
-      "required_pages_per_second_upper_estimate": 0.0017295255927212178
+      "required_token_positions_per_second": 28741.748427876373,
+      "required_pages_per_second_upper_estimate": 0.0017230239321424272
     },
     {
       "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": 28850.20257597789,
-      "required_pages_per_second_upper_estimate": 0.00224297635165221
+      "required_token_positions_per_second": 28741.748427876373,
+      "required_pages_per_second_upper_estimate": 0.0022345445186766994
     },
     {
       "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": 28850.20257597789,
-      "required_pages_per_second_upper_estimate": 0.3700594217105718
+      "required_token_positions_per_second": 28741.748427876373,
+      "required_pages_per_second_upper_estimate": 0.3686682883477171
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.060914860273426186,
-    "producer_p95_seconds": 0.13435649699999885,
-    "gpu_update_seconds": 0.9583294927371184,
-    "producer_batches_per_second": 16.41635547568095,
-    "gpu_batches_per_second": 1.0434824427075338,
-    "producer_headroom": 15.732277615601543,
+    "producer_batch_seconds": 0.4400104933984217,
+    "producer_p95_seconds": 0.144557610999982,
+    "gpu_update_seconds": 0.9619456543981313,
+    "producer_batches_per_second": 2.2726730725817434,
+    "gpu_batches_per_second": 1.03955976663326,
+    "producer_headroom": 2.186187986037657,
     "sustainable": true,
-    "buffer_seconds": 60.37475804243846,
+    "buffer_seconds": 60.602576227082274,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332477.80291559,
-    "upload_bytes_per_second": 4746298.913782841,
+    "generation_bytes_per_second": 297905.8677084344,
+    "upload_bytes_per_second": 5062848.150281854,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.34199681854987,
+    "estimated_full_shards_per_hour": 63.92366431655668,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 7055686,
       "pages": 40,
       "shards": 2
+    },
+    {
+      "seconds": 1.0181373959999291,
+      "bytes": 10411775,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 398600, "loss": 0.5956971049308777, "lr": 0.0003, "grad_norm": 0.14707166224125492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58343660275, "tokens_per_second": 151360.93894249553, "data_wait_s": 8.527000090907677e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 398700, "loss": 0.6194034218788147, "lr": 0.0003, "grad_norm": 0.14603193163626443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58358359562, "tokens_per_second": 152771.0292765205, "data_wait_s": 9.671000043454114e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 398800, "loss": 0.7059101462364197, "lr": 0.0003, "grad_norm": 0.30674263162538873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58373137537, "tokens_per_second": 135815.1271643288, "data_wait_s": 1.0689000191632658e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 398900, "loss": 0.6563550233840942, "lr": 0.0003, "grad_norm": 0.20655283200777552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58387974108, "tokens_per_second": 154349.77003529714, "data_wait_s": 8.372000138479052e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b9d7d96aa7eab8fce680c3abc9a93313f9035f324a17a4f87d16383e823f2b8
-size 36171
+oid sha256:f8d2e14bc8cf0bf69a641564eae209c2ea771afd304b86b3a087e83b8e466804
+size 38289