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": 76105.76941737354,
-      "required_pages_per_second_upper_estimate": 0.0052651737180082646
+      "required_token_positions_per_second": 77405.67765891201,
+      "required_pages_per_second_upper_estimate": 0.005355104386360568
     },
     {
       "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": 9513.221177171692,
-      "required_pages_per_second_upper_estimate": 0.00398298704823338
+      "required_token_positions_per_second": 9675.709707364002,
+      "required_pages_per_second_upper_estimate": 0.004051017602678542
     },
     {
       "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": 19026.442354343384,
-      "required_pages_per_second_upper_estimate": 0.003962706181382876
+      "required_token_positions_per_second": 19351.419414728003,
+      "required_pages_per_second_upper_estimate": 0.004030390332839587
     },
     {
       "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": 28539.663531515074,
-      "required_pages_per_second_upper_estimate": 0.0017109092511713377
+      "required_token_positions_per_second": 29027.129122092007,
+      "required_pages_per_second_upper_estimate": 0.0017401320690095653
     },
     {
       "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": 28539.663531515074,
-      "required_pages_per_second_upper_estimate": 0.002218833306862128
+      "required_token_positions_per_second": 29027.129122092007,
+      "required_pages_per_second_upper_estimate": 0.002256731612394951
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28539.663531515074,
-      "required_pages_per_second_upper_estimate": 0.3660761602790507
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29027.129122092007,
+      "required_pages_per_second_upper_estimate": 0.37232884547519923
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1008662234296196,
-    "producer_p95_seconds": 0.2910494249990734,
-    "gpu_update_seconds": 0.9687570412128212,
-    "producer_batches_per_second": 9.914121556238891,
-    "gpu_batches_per_second": 1.0322505617590811,
-    "producer_headroom": 9.60437506504624,
+    "producer_batch_seconds": 0.10933008868764205,
+    "producer_p95_seconds": 0.2867478759981168,
+    "gpu_update_seconds": 0.9524882699804309,
+    "producer_batches_per_second": 9.146612904129414,
+    "gpu_batches_per_second": 1.0498816956775177,
+    "producer_headroom": 8.71204150123491,
     "sustainable": true,
-    "buffer_seconds": 60.062936555194916,
+    "buffer_seconds": 60.00676100876715,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371269.35917070275,
-    "upload_bytes_per_second": 4902520.230367588,
+    "generation_bytes_per_second": 371138.67335138744,
+    "upload_bytes_per_second": 4819197.454597539,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.66576176968395,
+    "estimated_full_shards_per_hour": 79.63771963506906,
     "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": 1.3777095789992018,
-      "bytes": 11141641,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 3.434384232998127,
       "bytes": 5971898,
@@ -562,6 +556,12 @@
       "bytes": 5732876,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.102688891998696,
+      "bytes": 2347731,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 909800, "loss": 0.5249332785606384, "lr": 3e-05, "grad_norm": 0.3468487905408248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133168760721, "tokens_per_second": 151445.54722398162, "data_wait_s": 1.8576000002212822e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 909900, "loss": 0.5134099125862122, "lr": 3e-05, "grad_norm": 0.33801191589987484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133183188531, "tokens_per_second": 149716.73789439353, "data_wait_s": 8.429000445175916e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 910000, "loss": 0.4874023199081421, "lr": 3e-05, "grad_norm": 0.31720257835153753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133197845917, "tokens_per_second": 152953.92772428304, "data_wait_s": 1.6943999071372673e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 910100, "loss": 0.5183731913566589, "lr": 3e-05, "grad_norm": 0.3192280041209559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133212291168, "tokens_per_second": 139551.42774045252, "data_wait_s": 1.3638000382343307e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f9be0c6802983687df6d8772eba1eb0797ec79b8a792391db132c5c24ae143b
-size 533580
+oid sha256:7d5d26d340ae3e8aa477a03883c7cf8434d00f61fd152b38fe69daad59a76c3e
+size 535698