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": 77464.14238602118,
-      "required_pages_per_second_upper_estimate": 0.005359149111838837
+      "required_token_positions_per_second": 77330.94688401867,
+      "required_pages_per_second_upper_estimate": 0.005349934337954159
     },
     {
       "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": 9683.017798252647,
-      "required_pages_per_second_upper_estimate": 0.004054077347723326
+      "required_token_positions_per_second": 9666.368360502334,
+      "required_pages_per_second_upper_estimate": 0.004047106575817067
     },
     {
       "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": 19366.035596505295,
-      "required_pages_per_second_upper_estimate": 0.0040334344980491625
+      "required_token_positions_per_second": 19332.73672100467,
+      "required_pages_per_second_upper_estimate": 0.004026499220432782
     },
     {
       "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": 29049.05339475794,
-      "required_pages_per_second_upper_estimate": 0.0017414463956794619
+      "required_token_positions_per_second": 28999.105081507,
+      "required_pages_per_second_upper_estimate": 0.0017384520705667266
     },
     {
       "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": 29049.05339475794,
-      "required_pages_per_second_upper_estimate": 0.002258436128160043
+      "required_token_positions_per_second": 28999.105081507,
+      "required_pages_per_second_upper_estimate": 0.002254552866504194
     },
     {
       "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": 29049.05339475794,
-      "required_pages_per_second_upper_estimate": 0.3726100665045079
+      "required_token_positions_per_second": 28999.105081507,
+      "required_pages_per_second_upper_estimate": 0.3719693831724452
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08307342785928995,
-    "producer_p95_seconds": 0.1727394009994896,
-    "gpu_update_seconds": 0.9517693958657266,
-    "producer_batches_per_second": 12.03754348133802,
-    "gpu_batches_per_second": 1.0506746742895667,
-    "producer_headroom": 11.456965486940502,
+    "producer_batch_seconds": 0.082129281851536,
+    "producer_p95_seconds": 0.16981854200003,
+    "gpu_update_seconds": 0.9534087318312242,
+    "producer_batches_per_second": 12.175925290661697,
+    "gpu_batches_per_second": 1.0488680946725624,
+    "producer_headroom": 11.608633490241498,
     "sustainable": true,
-    "buffer_seconds": 60.913241335406504,
+    "buffer_seconds": 60.064750105367125,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332735.0850280428,
-    "upload_bytes_per_second": 4042742.2569695213,
+    "generation_bytes_per_second": 333821.0839538552,
+    "upload_bytes_per_second": 4089756.975589632,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.39720357066119,
+    "estimated_full_shards_per_hour": 71.63023365937941,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 4979236,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2744999360002112,
+      "bytes": 7432872,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 932500, "loss": 0.5872251987457275, "lr": 3e-05, "grad_norm": 0.4770336865020004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136491922411, "tokens_per_second": 149016.72008743306, "data_wait_s": 1.1174999599461444e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 932600, "loss": 0.5132953524589539, "lr": 3e-05, "grad_norm": 0.3311236421500409, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136506554592, "tokens_per_second": 152716.51894423802, "data_wait_s": 1.4995000128692482e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 932700, "loss": 0.5129926800727844, "lr": 3e-05, "grad_norm": 0.3344476085776876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136520966178, "tokens_per_second": 150316.800938626, "data_wait_s": 1.321299987466773e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 932800, "loss": 0.5275412797927856, "lr": 3e-05, "grad_norm": 0.35713417561118094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136535596977, "tokens_per_second": 152941.3422571803, "data_wait_s": 1.138099923991831e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:66348415964c9598b556094993506a081737601a290d0ea4e4d450543c0d3d51
-size 164034
+oid sha256:1b02aafafd09611610d7eb34f9e1a737bd433c876ecad25c9d8610468a793e81
+size 166152