Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75241.41449879439,
-      "required_pages_per_second_upper_estimate": 0.006521735111133893
+      "required_token_positions_per_second": 75212.49546489882,
+      "required_pages_per_second_upper_estimate": 0.0065192284825704465
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9405.176812349298,
-      "required_pages_per_second_upper_estimate": 0.003751865650370711
+      "required_token_positions_per_second": 9401.561933112353,
+      "required_pages_per_second_upper_estimate": 0.0037504236209958326
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18810.353624698597,
-      "required_pages_per_second_upper_estimate": 0.004351444260157019
+      "required_token_positions_per_second": 18803.123866224705,
+      "required_pages_per_second_upper_estimate": 0.004349771782773487
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28215.530437047895,
-      "required_pages_per_second_upper_estimate": 0.001718177768168796
+      "required_token_positions_per_second": 28204.68579933706,
+      "required_pages_per_second_upper_estimate": 0.0017175173866296229
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28215.530437047895,
-      "required_pages_per_second_upper_estimate": 0.0021721096386311768
+      "required_token_positions_per_second": 28204.68579933706,
+      "required_pages_per_second_upper_estimate": 0.002171274788400318
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28215.530437047895,
-      "required_pages_per_second_upper_estimate": 0.3503424567223499
+      "required_token_positions_per_second": 28204.68579933706,
+      "required_pages_per_second_upper_estimate": 0.350207802616649
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.06761172991411968,
-    "producer_p95_seconds": 0.26857647799988627,
-    "gpu_update_seconds": 0.9798858845374493,
-    "producer_batches_per_second": 14.79033299799012,
-    "gpu_batches_per_second": 1.0205269978677625,
-    "producer_headroom": 14.492838532338974,
+    "producer_batch_seconds": 0.0867094905778174,
+    "producer_p95_seconds": 0.2796822970012727,
+    "gpu_update_seconds": 0.9802626484372982,
+    "producer_batches_per_second": 11.532762945972452,
+    "gpu_batches_per_second": 1.0201347583672258,
+    "producer_headroom": 11.305136749218493,
     "sustainable": true,
-    "buffer_seconds": 60.75292484132186,
+    "buffer_seconds": 60.77628420311249,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 157,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 540098.4572457317,
-    "upload_bytes_per_second": 5780854.552689143,
+    "generation_bytes_per_second": 565471.0668516669,
+    "upload_bytes_per_second": 5704649.266604119,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.89255607632603,
+    "estimated_full_shards_per_hour": 60.66846372681859,
     "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": 0.9899479589985276,
-      "bytes": 6233203,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.8259697120010969,
       "bytes": 2261871,
@@ -562,6 +556,12 @@
       "bytes": 11844982,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.079595678998885,
+      "bytes": 1118845,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 372300, "loss": 0.5817138999700546, "lr": 0.0003, "grad_norm": 0.17083567174265302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54493951271, "tokens_per_second": 150660.22467692784, "data_wait_s": 1.0944000678136945e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 372400, "loss": 0.5703586395829916, "lr": 0.0003, "grad_norm": 0.1545383032913337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54508489145, "tokens_per_second": 148005.0871590917, "data_wait_s": 1.5253001038217917e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 372500, "loss": 0.5488224159926176, "lr": 0.0003, "grad_norm": 0.1557408398169736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54523095281, "tokens_per_second": 148776.12141943202, "data_wait_s": 9.688999853096902e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 372600, "loss": 0.5903936438262463, "lr": 0.0003, "grad_norm": 0.15902455622511727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54537546930, "tokens_per_second": 147442.5573604477, "data_wait_s": 2.0613999367924407e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0feeb501f9203087eea30a24a20ea1d1b581568cedee9f8ff6137c15044c25c
-size 373164
+oid sha256:2144e0f1e202b537efea5b16b2894b06bab4dfdc144a7ac1383baaf5f0d2b5fb
+size 375282