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": 77690.35416741742,
-      "required_pages_per_second_upper_estimate": 0.005374798967759477
+      "required_token_positions_per_second": 77487.32080631121,
+      "required_pages_per_second_upper_estimate": 0.005360752648735846
     },
     {
       "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": 9711.294270927177,
-      "required_pages_per_second_upper_estimate": 0.004065916116352257
+      "required_token_positions_per_second": 9685.915100788901,
+      "required_pages_per_second_upper_estimate": 0.0040552903877926995
     },
     {
       "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": 19422.588541854355,
-      "required_pages_per_second_upper_estimate": 0.00404521298516391
+      "required_token_positions_per_second": 19371.830201577803,
+      "required_pages_per_second_upper_estimate": 0.004034641361471754
     },
     {
       "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": 29133.88281278153,
-      "required_pages_per_second_upper_estimate": 0.0017465317897629546
+      "required_token_positions_per_second": 29057.7453023667,
+      "required_pages_per_second_upper_estimate": 0.0017419674622688327
     },
     {
       "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": 29133.88281278153,
-      "required_pages_per_second_upper_estimate": 0.0022650312422862005
+      "required_token_positions_per_second": 29057.7453023667,
+      "required_pages_per_second_upper_estimate": 0.002259111885744963
     },
     {
       "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": 29133.88281278153,
-      "required_pages_per_second_upper_estimate": 0.3736981671961818
+      "required_token_positions_per_second": 29057.7453023667,
+      "required_pages_per_second_upper_estimate": 0.3727215569626698
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07620882570320475,
-    "producer_p95_seconds": 0.20124244500038913,
-    "gpu_update_seconds": 0.9489981193948632,
-    "producer_batches_per_second": 13.121839770822604,
-    "gpu_batches_per_second": 1.0537428679391467,
-    "producer_headroom": 12.452601265511374,
+    "producer_batch_seconds": 0.09813096410164235,
+    "producer_p95_seconds": 0.20736018299976422,
+    "gpu_update_seconds": 0.9514846975325412,
+    "producer_batches_per_second": 10.19046341951983,
+    "gpu_batches_per_second": 1.0509890517349068,
+    "producer_headroom": 9.696070004438251,
     "sustainable": true,
-    "buffer_seconds": 60.73587964127125,
+    "buffer_seconds": 60.895020642082635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416405.05368189554,
-    "upload_bytes_per_second": 5283058.194142973,
+    "generation_bytes_per_second": 427102.26453449414,
+    "upload_bytes_per_second": 5350317.881374578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.35083110659265,
+    "estimated_full_shards_per_hour": 91.64620353723639,
     "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": 1.5451833110000734,
-      "bytes": 7247709,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.4297130429995377,
       "bytes": 4940477,
@@ -562,6 +556,12 @@
       "bytes": 2607815,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1019334159991558,
+      "bytes": 10289866,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -129,3 +129,4 @@
 {"step": 657900, "loss": 0.5477266311645508, "lr": 0.0001, "grad_norm": 0.20004618513616076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96298802646, "tokens_per_second": 153721.23076655628, "data_wait_s": 1.2876000255346298e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 658000, "loss": 0.5574698448181152, "lr": 0.0001, "grad_norm": 0.18377159777841284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96313493310, "tokens_per_second": 153809.93734006878, "data_wait_s": 9.886000043479726e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 658100, "loss": 0.5140765905380249, "lr": 0.0001, "grad_norm": 0.2048446612348188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96328169625, "tokens_per_second": 153348.09189666715, "data_wait_s": 8.286000593216158e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 658200, "loss": 0.5287087559700012, "lr": 0.0001, "grad_norm": 0.18828733658335056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96342884025, "tokens_per_second": 152876.57498918302, "data_wait_s": 1.3075001334073022e-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:b6e9dcc6b9d50ddd0ee7ea6067810d276dc2d283fb9ec25527c131a81a433296
-size 279189
+oid sha256:ee47ce3f35c26842ce19c98331a91757164092d79809ded25041ac1677b721ab
+size 281307