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": 77365.63508635426,
-      "required_pages_per_second_upper_estimate": 0.005352334148279454
+      "required_token_positions_per_second": 77156.6150675723,
+      "required_pages_per_second_upper_estimate": 0.005337873658386855
     },
     {
       "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": 9670.704385794283,
-      "required_pages_per_second_upper_estimate": 0.004048921979060301
+      "required_token_positions_per_second": 9644.576883446538,
+      "required_pages_per_second_upper_estimate": 0.004037982939431592
     },
     {
       "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": 19341.408771588565,
-      "required_pages_per_second_upper_estimate": 0.004028305379872054
+      "required_token_positions_per_second": 19289.153766893076,
+      "required_pages_per_second_upper_estimate": 0.004017422040451128
     },
     {
       "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": 29012.113157382846,
-      "required_pages_per_second_upper_estimate": 0.0017392318848533017
+      "required_token_positions_per_second": 28933.730650339614,
+      "required_pages_per_second_upper_estimate": 0.0017345329732392182
     },
     {
       "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": 29012.113157382846,
-      "required_pages_per_second_upper_estimate": 0.002255564187187062
+      "required_token_positions_per_second": 28933.730650339614,
+      "required_pages_per_second_upper_estimate": 0.0022494702920326628
     },
     {
       "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": 29012.113157382846,
-      "required_pages_per_second_upper_estimate": 0.37213623680279684
+      "required_token_positions_per_second": 28933.730650339614,
+      "required_pages_per_second_upper_estimate": 0.3711308301630253
     }
   ],
   "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": 0.08169381431264355,
+    "producer_batch_seconds": 0.37896600327334795,
     "producer_p95_seconds": 0.23710926200146787,
-    "gpu_update_seconds": 0.9529812547613163,
-    "producer_batches_per_second": 12.240828860958601,
-    "gpu_batches_per_second": 1.0493385835280253,
-    "producer_headroom": 11.665280447234862,
+    "gpu_update_seconds": 0.9555629149286865,
+    "producer_batches_per_second": 2.6387591271048674,
+    "gpu_batches_per_second": 1.0465035680823065,
+    "producer_headroom": 2.5215003632910036,
     "sustainable": true,
-    "buffer_seconds": 60.03781904996293,
+    "buffer_seconds": 60.20046364050725,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491839.2267922314,
+    "generation_bytes_per_second": 441218.51279274194,
     "upload_bytes_per_second": 5219095.352802215,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.53724863839346,
+    "estimated_full_shards_per_hour": 94.6752218040151,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 578500, "loss": 0.5562648177146912, "lr": 0.0003, "grad_norm": 0.14849490008911154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84675307075, "tokens_per_second": 151294.98359596176, "data_wait_s": 1.0092000593431294e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 578600, "loss": 0.5609111189842224, "lr": 0.0003, "grad_norm": 0.17871318168536346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84689857947, "tokens_per_second": 151580.5485364575, "data_wait_s": 9.054001566255465e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 578700, "loss": 0.6683929562568665, "lr": 0.0003, "grad_norm": 0.18126979748246194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84704366090, "tokens_per_second": 150255.4647987151, "data_wait_s": 1.0960000508930534e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 578800, "loss": 0.709212064743042, "lr": 0.0003, "grad_norm": 0.16372910982491218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84719087562, "tokens_per_second": 154262.92936551804, "data_wait_s": 1.0092000593431294e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16a9a3b2304e4c067e933f3f489d0d29060cdf990bff0202a9e24ca5189e3baa
-size 460785
+oid sha256:0670c4220e2ed2c94e2a2d67778758a016cc20a35a92d0807e0a037b185de2ad
+size 462903