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": 77431.50879263022,
-      "required_pages_per_second_upper_estimate": 0.005356891444127682
+      "required_token_positions_per_second": 77162.61061882264,
+      "required_pages_per_second_upper_estimate": 0.005338288444533949
     },
     {
       "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": 9678.938599078778,
-      "required_pages_per_second_upper_estimate": 0.004052369472212593
+      "required_token_positions_per_second": 9645.32632735283,
+      "required_pages_per_second_upper_estimate": 0.0040382967159449884
     },
     {
       "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": 19357.877198157556,
-      "required_pages_per_second_upper_estimate": 0.004031735318824761
+      "required_token_positions_per_second": 19290.65265470566,
+      "required_pages_per_second_upper_estimate": 0.004017734219254159
     },
     {
       "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": 29036.81579723633,
-      "required_pages_per_second_upper_estimate": 0.0017407127703937709
+      "required_token_positions_per_second": 28935.97898205849,
+      "required_pages_per_second_upper_estimate": 0.0017346677572927611
     },
     {
       "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": 29036.81579723633,
-      "required_pages_per_second_upper_estimate": 0.002257484708780241
+      "required_token_positions_per_second": 28935.97898205849,
+      "required_pages_per_second_upper_estimate": 0.0022496450899344394
     },
     {
       "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": 29036.81579723633,
-      "required_pages_per_second_upper_estimate": 0.3724530957432092
+      "required_token_positions_per_second": 28935.97898205849,
+      "required_pages_per_second_upper_estimate": 0.37115966934824446
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0793044515468253,
-    "producer_p95_seconds": 0.18340161599917337,
-    "gpu_update_seconds": 0.9521705201102486,
-    "producer_batches_per_second": 12.609632630893993,
-    "gpu_batches_per_second": 1.0502320528514297,
-    "producer_headroom": 12.006520460557496,
+    "producer_batch_seconds": 0.07075502796097766,
+    "producer_p95_seconds": 0.1850161170004867,
+    "gpu_update_seconds": 0.9554886674870379,
+    "producer_batches_per_second": 14.133271215036666,
+    "gpu_batches_per_second": 1.0465848879506108,
+    "producer_headroom": 13.504180480488293,
     "sustainable": true,
-    "buffer_seconds": 60.93891328705591,
+    "buffer_seconds": 60.195786051683385,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375961.03283960285,
-    "upload_bytes_per_second": 4843615.443185026,
+    "generation_bytes_per_second": 327899.80789341935,
+    "upload_bytes_per_second": 4865565.867784171,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.67248572245659,
+    "estimated_full_shards_per_hour": 70.35966565706191,
     "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": [
@@ -514,6 +514,12 @@
       "bytes": 11593993,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 2.0578567419997853,
+      "bytes": 11618530,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -66,3 +66,4 @@
 {"step": 403700, "loss": 0.5862820148468018, "lr": 0.0003, "grad_norm": 0.16875477108175826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59091191852, "tokens_per_second": 149825.8917697164, "data_wait_s": 1.5505000192206353e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 403800, "loss": 0.5630764961242676, "lr": 0.0003, "grad_norm": 0.16310916483965648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59105768262, "tokens_per_second": 152156.5114509549, "data_wait_s": 9.673999556980561e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 403900, "loss": 0.563493013381958, "lr": 0.0003, "grad_norm": 0.157540287331852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59120453389, "tokens_per_second": 122755.2944953602, "data_wait_s": 9.309999768447597e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 404000, "loss": 0.576484203338623, "lr": 0.0003, "grad_norm": 0.17450364148191946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59135043101, "tokens_per_second": 151872.57447063093, "data_wait_s": 1.0945999747491442e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fa3e56ef5e1111ff2cdf0836e0576b2868ae11997794f2f4bf338b3adc5fe8ec
-size 144972
+oid sha256:0df321da6bce39f00e6d1f6338473c0141954b897d3af45e763d7275410f4d07
+size 147090