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": 76687.06380312912,
-      "required_pages_per_second_upper_estimate": 0.005305389012403637
+      "required_token_positions_per_second": 77203.9666896981,
+      "required_pages_per_second_upper_estimate": 0.005341149553476419
     },
     {
       "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": 9585.88297539114,
-      "required_pages_per_second_upper_estimate": 0.004013409025796972
+      "required_token_positions_per_second": 9650.495836212263,
+      "required_pages_per_second_upper_estimate": 0.004040461081352812
     },
     {
       "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": 19171.76595078228,
-      "required_pages_per_second_upper_estimate": 0.003992973254080138
+      "required_token_positions_per_second": 19300.991672424527,
+      "required_pages_per_second_upper_estimate": 0.004019887563986767
     },
     {
       "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": 28757.648926173417,
-      "required_pages_per_second_upper_estimate": 0.00172397714273668
+      "required_token_positions_per_second": 28951.48750863679,
+      "required_pages_per_second_upper_estimate": 0.0017355974697809812
     },
     {
       "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": 28757.648926173417,
-      "required_pages_per_second_upper_estimate": 0.002235780712480394
+      "required_token_positions_per_second": 28951.48750863679,
+      "required_pages_per_second_upper_estimate": 0.002250850809661103
     },
     {
       "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": 28757.648926173417,
-      "required_pages_per_second_upper_estimate": 0.3688722428672467
+      "required_token_positions_per_second": 28951.48750863679,
+      "required_pages_per_second_upper_estimate": 0.3713585960754325
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11509934371878217,
-    "producer_p95_seconds": 0.23064451400023245,
-    "gpu_update_seconds": 0.9614137814596003,
-    "producer_batches_per_second": 8.688146845070307,
-    "gpu_batches_per_second": 1.0401348714617122,
-    "producer_headroom": 8.35290411219534,
+    "producer_batch_seconds": 0.06164197364060442,
+    "producer_p95_seconds": 0.20303044599859277,
+    "gpu_update_seconds": 0.9549768381245373,
+    "producer_batches_per_second": 16.222712235503217,
+    "gpu_batches_per_second": 1.0471458155612265,
+    "producer_headroom": 15.492314436465104,
     "sustainable": true,
-    "buffer_seconds": 60.56906823195482,
+    "buffer_seconds": 60.16354080184585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 533035.2450728753,
-    "upload_bytes_per_second": 2196191.4750022837,
+    "generation_bytes_per_second": 533304.8141429316,
+    "upload_bytes_per_second": 2145941.4062687345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.37695516719526,
+    "estimated_full_shards_per_hour": 114.43479841438257,
     "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": 1.5570587419997537,
-      "bytes": 16077242,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.4845605279997471,
       "bytes": 2728370,
@@ -562,6 +556,12 @@
       "bytes": 2537038,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5384773589994438,
+      "bytes": 6037444,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -151,3 +151,4 @@
 {"step": 660100, "loss": 0.5888491868972778, "lr": 0.0001, "grad_norm": 0.1944544811202886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96621069918, "tokens_per_second": 131941.47359829125, "data_wait_s": 2.1578000087174587e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 660200, "loss": 0.5271904468536377, "lr": 0.0001, "grad_norm": 0.20451191649112574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96635743357, "tokens_per_second": 153173.1027056159, "data_wait_s": 1.3371000022743829e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 660300, "loss": 0.5828812718391418, "lr": 0.0001, "grad_norm": 0.21404818688318966, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96650522328, "tokens_per_second": 154709.3926726605, "data_wait_s": 8.66899972606916e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 660400, "loss": 0.5669741630554199, "lr": 0.0001, "grad_norm": 0.20515818367769723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96665258997, "tokens_per_second": 153529.86450996198, "data_wait_s": 1.4510000255540945e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55e40eb40ebc4921b432adb72ffb2b78b1e2f239427d18ce8f7200c30e761a00
-size 326568
+oid sha256:b8b2324f417e18174bdac883abe06a63867f3936544f68b6198a2bede11f9c81
+size 328686