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": 76528.43666681554,
-      "required_pages_per_second_upper_estimate": 0.005294414819047798
+      "required_token_positions_per_second": 77130.33187816339,
+      "required_pages_per_second_upper_estimate": 0.005336055326358145
     },
     {
       "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": 9566.054583351943,
-      "required_pages_per_second_upper_estimate": 0.004005107292114071
+      "required_token_positions_per_second": 9641.291484770423,
+      "required_pages_per_second_upper_estimate": 0.004036607411612829
     },
     {
       "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": 19132.109166703885,
-      "required_pages_per_second_upper_estimate": 0.003984713791776334
+      "required_token_positions_per_second": 19282.582969540847,
+      "required_pages_per_second_upper_estimate": 0.004016053516646237
     },
     {
       "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": 28698.163750055824,
-      "required_pages_per_second_upper_estimate": 0.0017204110972570351
+      "required_token_positions_per_second": 28923.874454311266,
+      "required_pages_per_second_upper_estimate": 0.0017339421093368588
     },
     {
       "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": 28698.163750055824,
-      "required_pages_per_second_upper_estimate": 0.002231156001684889
+      "required_token_positions_per_second": 28923.874454311266,
+      "required_pages_per_second_upper_estimate": 0.0022487040161442836
     },
     {
       "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": 28698.163750055824,
-      "required_pages_per_second_upper_estimate": 0.36810923089821607
+      "required_token_positions_per_second": 28923.874454311266,
+      "required_pages_per_second_upper_estimate": 0.3710044054631324
     }
   ],
   "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": 0.1191363028205501,
-    "producer_p95_seconds": 0.3289518569945358,
-    "gpu_update_seconds": 0.9634065872924088,
-    "producer_batches_per_second": 8.393747131017294,
-    "gpu_batches_per_second": 1.0379833532282923,
-    "producer_headroom": 8.08659127808882,
+    "producer_batch_seconds": 0.106630585258074,
+    "producer_p95_seconds": 0.32927844299410935,
+    "gpu_update_seconds": 0.9558885357379536,
+    "producer_batches_per_second": 9.378172290621285,
+    "gpu_batches_per_second": 1.0461470795106795,
+    "producer_headroom": 8.964487378780229,
     "sustainable": true,
-    "buffer_seconds": 60.69461499942175,
+    "buffer_seconds": 60.22097775149108,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 645988.2628555205,
-    "upload_bytes_per_second": 5431514.106085738,
+    "generation_bytes_per_second": 643144.7772215593,
+    "upload_bytes_per_second": 5362182.808867596,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.30702168583494,
+    "estimated_full_shards_per_hour": 69.0019487737898,
     "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": 1.1143527310014179,
-      "bytes": 6164100,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.3281419349987118,
       "bytes": 11547797,
@@ -562,6 +556,12 @@
       "bytes": 10238885,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.4429230239984463,
+      "bytes": 2385665,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -355,3 +355,4 @@
 {"step": 680500, "loss": 0.592078447341919, "lr": 0.0001, "grad_norm": 0.2525708030391718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99607513653, "tokens_per_second": 150977.59807706944, "data_wait_s": 1.757699646987021e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 680600, "loss": 0.5751513838768005, "lr": 0.0001, "grad_norm": 0.20857990913504343, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99622064215, "tokens_per_second": 151103.80003626796, "data_wait_s": 1.5407000319100916e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 680700, "loss": 0.5364224910736084, "lr": 0.0001, "grad_norm": 0.2100102845041653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99636574961, "tokens_per_second": 151614.97339502323, "data_wait_s": 1.017199974739924e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 680800, "loss": 0.5573974251747131, "lr": 0.0001, "grad_norm": 0.1978600324144116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99651232369, "tokens_per_second": 152168.42304824668, "data_wait_s": 1.774200063664466e-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:defcd045026d092c94f635aa8f42d57a3c08bf0138c0ccf6bb4f96aef1bc5604
-size 761772
+oid sha256:eb704e5f034d8c0be6cc88eef2e070b0c0f38fa5777df764e174dc8d84a77d50
+size 763890