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": 76926.56067631026,
-      "required_pages_per_second_upper_estimate": 0.005321957961799616
+      "required_token_positions_per_second": 76725.1074481779,
+      "required_pages_per_second_upper_estimate": 0.005308020960041531
     },
     {
       "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": 9615.820084538782,
-      "required_pages_per_second_upper_estimate": 0.004025943068239162
+      "required_token_positions_per_second": 9590.638431022238,
+      "required_pages_per_second_upper_estimate": 0.004015400035764507
     },
     {
       "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": 19231.640169077564,
-      "required_pages_per_second_upper_estimate": 0.004005443474761727
+      "required_token_positions_per_second": 19181.276862044477,
+      "required_pages_per_second_upper_estimate": 0.003994954126076458
     },
     {
       "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": 28847.460253616344,
-      "required_pages_per_second_upper_estimate": 0.0017293611946829267
+      "required_token_positions_per_second": 28771.91529306672,
+      "required_pages_per_second_upper_estimate": 0.0017248323896484525
     },
     {
       "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": 28847.460253616344,
-      "required_pages_per_second_upper_estimate": 0.002242763148150801
+      "required_token_positions_per_second": 28771.91529306672,
+      "required_pages_per_second_upper_estimate": 0.002236889859755232
     },
     {
       "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": 28847.460253616344,
-      "required_pages_per_second_upper_estimate": 0.37002424614379426
+      "required_token_positions_per_second": 28771.91529306672,
+      "required_pages_per_second_upper_estimate": 0.36905523650372263
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0946611663437551,
-    "producer_p95_seconds": 0.1866953190001368,
-    "gpu_update_seconds": 0.9584205942890248,
-    "producer_batches_per_second": 10.563994070900977,
-    "gpu_batches_per_second": 1.0433832557008227,
-    "producer_headroom": 10.12474947549865,
+    "producer_batch_seconds": 0.05429653879679108,
+    "producer_p95_seconds": 0.18611543899896787,
+    "gpu_update_seconds": 0.9609370706948539,
+    "producer_batches_per_second": 18.417380226437192,
+    "gpu_batches_per_second": 1.040650871421684,
+    "producer_headroom": 17.69794340466588,
     "sustainable": true,
-    "buffer_seconds": 60.380497440208565,
+    "buffer_seconds": 60.539035453775796,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 304363.31942024117,
-    "upload_bytes_per_second": 5018664.44489128,
+    "generation_bytes_per_second": 321114.00341084995,
+    "upload_bytes_per_second": 5003881.846550181,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.30928313212802,
+    "estimated_full_shards_per_hour": 68.90358998054623,
     "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": 1.1527475369998683,
-      "bytes": 8325445,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.049984759000381,
       "bytes": 5490081,
@@ -562,6 +556,12 @@
       "bytes": 1155225,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.29576844300027,
+      "bytes": 7786718,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -107,3 +107,4 @@
 {"step": 407800, "loss": 0.5638266205787659, "lr": 0.0003, "grad_norm": 0.1509126354770479, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59691261272, "tokens_per_second": 152007.03551981243, "data_wait_s": 1.0596999345580116e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 407900, "loss": 0.5310512781143188, "lr": 0.0003, "grad_norm": 0.13714952542781725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59705922815, "tokens_per_second": 153166.61718350233, "data_wait_s": 1.734999932523351e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 408000, "loss": 0.5376939177513123, "lr": 0.0003, "grad_norm": 0.16437156780921008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59720455009, "tokens_per_second": 151925.93550526028, "data_wait_s": 1.2930000593769364e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 408100, "loss": 0.5757578611373901, "lr": 0.0003, "grad_norm": 0.15850202302499988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59735165744, "tokens_per_second": 153632.57055071363, "data_wait_s": 8.314998922287486e-06, "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:c4b0f1813da5ff786042b9bfb443c6ce5e865cf15589f788f97c42ee301e9b07
-size 232593
+oid sha256:f5e6b7db2a8d654a61e2ce46895d6c531dd7cec64b5e5b875809bd008abc2071
+size 234711