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": 76858.6171460907,
-      "required_pages_per_second_upper_estimate": 0.005317257470728317
+      "required_token_positions_per_second": 77127.30061390494,
+      "required_pages_per_second_upper_estimate": 0.005335845616463244
     },
     {
       "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": 9607.327143261338,
-      "required_pages_per_second_upper_estimate": 0.004022387251079077
+      "required_token_positions_per_second": 9640.912576738117,
+      "required_pages_per_second_upper_estimate": 0.004036448770732202
     },
     {
       "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": 19214.654286522677,
-      "required_pages_per_second_upper_estimate": 0.004001905763373391
+      "required_token_positions_per_second": 19281.825153476235,
+      "required_pages_per_second_upper_estimate": 0.004015895683544933
     },
     {
       "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": 28821.981429784013,
-      "required_pages_per_second_upper_estimate": 0.0017278337780980899
+      "required_token_positions_per_second": 28922.737730214354,
+      "required_pages_per_second_upper_estimate": 0.0017338739644629261
     },
     {
       "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": 28821.981429784013,
-      "required_pages_per_second_upper_estimate": 0.0022407822816673394
+      "required_token_positions_per_second": 28922.737730214354,
+      "required_pages_per_second_upper_estimate": 0.0022486156408456697
     },
     {
       "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": 28821.981429784013,
-      "required_pages_per_second_upper_estimate": 0.36969743114870274
+      "required_token_positions_per_second": 28922.737730214354,
+      "required_pages_per_second_upper_estimate": 0.3709898247869365
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09381637764067818,
-    "producer_p95_seconds": 0.2846004519997223,
-    "gpu_update_seconds": 0.9592678444872341,
-    "producer_batches_per_second": 10.659119709674298,
-    "gpu_batches_per_second": 1.0424617125934612,
-    "producer_headroom": 10.224950788030657,
+    "producer_batch_seconds": 0.08810394916397968,
+    "producer_p95_seconds": 0.2701275009967503,
+    "gpu_update_seconds": 0.9559261041570526,
+    "producer_batches_per_second": 11.35022901344403,
+    "gpu_batches_per_second": 1.0461059653578686,
+    "producer_headroom": 10.849980202111897,
     "sustainable": true,
-    "buffer_seconds": 60.43387420269575,
+    "buffer_seconds": 60.223344561894315,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 389250.4300975164,
+    "generation_bytes_per_second": 406205.3270278003,
     "upload_bytes_per_second": 3893598.4107209784,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.52408101266974,
+    "estimated_full_shards_per_hour": 87.16220720410831,
     "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": [

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

@@ -233,3 +233,4 @@
 {"step": 668300, "loss": 0.583810567855835, "lr": 0.0001, "grad_norm": 0.19131607142650467, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97822025396, "tokens_per_second": 153182.6814426646, "data_wait_s": 8.918999810703099e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 668400, "loss": 0.5436054468154907, "lr": 0.0001, "grad_norm": 0.1865470383934079, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97836766563, "tokens_per_second": 153155.22006086787, "data_wait_s": 1.573100234963931e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 668500, "loss": 0.5445132851600647, "lr": 0.0001, "grad_norm": 0.19579207798574336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97851431435, "tokens_per_second": 152858.7811853542, "data_wait_s": 1.4663000911241397e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 668600, "loss": 0.5461285710334778, "lr": 0.0001, "grad_norm": 0.21074693561526772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97865898600, "tokens_per_second": 150306.41581491462, "data_wait_s": 1.622700074221939e-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 sha256:3e69b243e017076ed402940459c258c21d97b46091cd764a88bc58e079cdf0ef
-size 501027
+oid sha256:6fb70178a07e283c5e5fed29e022d8d40e9017c180ab61f3b49774051c0f0ba8
+size 503145