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": 76222.81712583438,
-      "required_pages_per_second_upper_estimate": 0.005273271350067686
+      "required_token_positions_per_second": 77139.22098056908,
+      "required_pages_per_second_upper_estimate": 0.00533667029508813
     },
     {
       "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": 9527.852140729297,
-      "required_pages_per_second_upper_estimate": 0.003989112727145688
+      "required_token_positions_per_second": 9642.402622571135,
+      "required_pages_per_second_upper_estimate": 0.004037072621806791
     },
     {
       "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": 19055.704281458595,
-      "required_pages_per_second_upper_estimate": 0.0039688006691120615
+      "required_token_positions_per_second": 19284.80524514227,
+      "required_pages_per_second_upper_estimate": 0.004016516358048666
     },
     {
       "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": 28583.556422187892,
-      "required_pages_per_second_upper_estimate": 0.0017135405629466096
+      "required_token_positions_per_second": 28927.2078677134,
+      "required_pages_per_second_upper_estimate": 0.0017341419423804892
     },
     {
       "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": 28583.556422187892,
-      "required_pages_per_second_upper_estimate": 0.0022222457860475173
+      "required_token_positions_per_second": 28927.2078677134,
+      "required_pages_per_second_upper_estimate": 0.0022489631743741636
     },
     {
       "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": 28583.556422187892,
-      "required_pages_per_second_upper_estimate": 0.3666391711520875
+      "required_token_positions_per_second": 28927.2078677134,
+      "required_pages_per_second_upper_estimate": 0.37104716291111456
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1025845575780977,
-    "producer_p95_seconds": 0.26451168499988853,
-    "gpu_update_seconds": 0.9672694185296806,
-    "producer_batches_per_second": 9.748055882959765,
-    "gpu_batches_per_second": 1.033838122908995,
-    "producer_headroom": 9.428996345705324,
+    "producer_batch_seconds": 0.09413797759353315,
+    "producer_p95_seconds": 0.27479024899730575,
+    "gpu_update_seconds": 0.9557783843652201,
+    "producer_batches_per_second": 10.622705368897742,
+    "gpu_batches_per_second": 1.0462676456782916,
+    "producer_headroom": 10.152952175072834,
     "sustainable": true,
-    "buffer_seconds": 60.93797336736988,
+    "buffer_seconds": 60.21403821500887,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 172,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 661627.8622252488,
+    "generation_bytes_per_second": 658674.4400519676,
     "upload_bytes_per_second": 3962966.409334982,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.98496866258668,
+    "estimated_full_shards_per_hour": 70.66810083946834,
     "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": [

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 823900, "loss": 0.5389016270637512, "lr": 3e-05, "grad_norm": 0.2831273121261325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120596101097, "tokens_per_second": 151948.56498266652, "data_wait_s": 1.629399775993079e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 824000, "loss": 0.5406765341758728, "lr": 3e-05, "grad_norm": 0.27961216585417376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120610787974, "tokens_per_second": 151657.88620235698, "data_wait_s": 8.94200275070034e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 824100, "loss": 0.5342902541160583, "lr": 3e-05, "grad_norm": 0.27888587842782986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120625466659, "tokens_per_second": 152007.76809514302, "data_wait_s": 1.367299773846753e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 824200, "loss": 0.49697187542915344, "lr": 3e-05, "grad_norm": 0.2743217369547591, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120639908563, "tokens_per_second": 147991.62353444542, "data_wait_s": 2.4401000700891018e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e428c7cc41ae08a08507cb153037b7e15ce02babfbd21ea6e445b8cd7ce46e0
-size 513735
+oid sha256:4dbf32a9d4b76c58d804c4ad6fa9b209150f778b6a925fa9d6f38ac035a37167
+size 515853