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": 76518.56638952837,
-      "required_pages_per_second_upper_estimate": 0.005293731970362873
+      "required_token_positions_per_second": 77125.70549117064,
+      "required_pages_per_second_upper_estimate": 0.005335735262171289
     },
     {
       "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": 9564.820798691046,
-      "required_pages_per_second_upper_estimate": 0.004004590732240907
+      "required_token_positions_per_second": 9640.71318639633,
+      "required_pages_per_second_upper_estimate": 0.004036365290159839
     },
     {
       "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": 19129.64159738209,
-      "required_pages_per_second_upper_estimate": 0.003984199862160783
+      "required_token_positions_per_second": 19281.42637279266,
+      "required_pages_per_second_upper_estimate": 0.004015812628045097
     },
     {
       "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": 28694.46239607314,
-      "required_pages_per_second_upper_estimate": 0.0017201892067374134
+      "required_token_positions_per_second": 28922.13955918899,
+      "required_pages_per_second_upper_estimate": 0.0017338381050233096
     },
     {
       "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": 28694.46239607314,
-      "required_pages_per_second_upper_estimate": 0.002230868237693271
+      "required_token_positions_per_second": 28922.13955918899,
+      "required_pages_per_second_upper_estimate": 0.0022485691356795238
     },
     {
       "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": 28694.46239607314,
-      "required_pages_per_second_upper_estimate": 0.36806175390353046
+      "required_token_positions_per_second": 28922.13955918899,
+      "required_pages_per_second_upper_estimate": 0.3709821520912891
     }
   ],
   "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": 0.44777965811664444,
+    "producer_batch_seconds": 0.2789654328821598,
     "producer_p95_seconds": 0.3613663889991585,
-    "gpu_update_seconds": 0.9635308589640506,
-    "producer_batches_per_second": 2.233241242369042,
-    "gpu_batches_per_second": 1.0378494790246362,
-    "producer_headroom": 2.1517968525337867,
+    "gpu_update_seconds": 0.9559458747309663,
+    "producer_batches_per_second": 3.5846735191109453,
+    "gpu_batches_per_second": 1.0460843301211296,
+    "producer_headroom": 3.426753862851444,
     "sustainable": true,
-    "buffer_seconds": 60.70244411473519,
+    "buffer_seconds": 60.224590108050876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340028.00613044115,
+    "generation_bytes_per_second": 337194.06884666724,
     "upload_bytes_per_second": 4500702.288970114,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.96209466872145,
+    "estimated_full_shards_per_hour": 72.35399769830717,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 645200, "loss": 0.5682960152626038, "lr": 0.0001, "grad_norm": 0.18092363405899495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94440612732, "tokens_per_second": 152416.5722103898, "data_wait_s": 1.475700264563784e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 645300, "loss": 0.5486984252929688, "lr": 0.0001, "grad_norm": 0.17951375196020838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94455290557, "tokens_per_second": 151987.93804769317, "data_wait_s": 2.100800338666886e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 645400, "loss": 0.5167580842971802, "lr": 0.0001, "grad_norm": 0.16668514075716886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94470019608, "tokens_per_second": 140380.28128611538, "data_wait_s": 7.217000529635698e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 645500, "loss": 0.5505189299583435, "lr": 0.0001, "grad_norm": 0.17340860427545055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94484547368, "tokens_per_second": 150862.3410131202, "data_wait_s": 1.1018004443030804e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e4deb79ab648b00781f995d72cc6b0e00a7d9533a7319e1fb5a4f654a57e8c6
-size 875592
+oid sha256:da51494136603433656b3ff2616edc4d3e9105564d96ca42e57005a8aa91d3dd
+size 877710