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": 77306.43268540049,
-      "required_pages_per_second_upper_estimate": 0.005348238388812976
+      "required_token_positions_per_second": 77135.08397927361,
+      "required_pages_per_second_upper_estimate": 0.005336384087739344
     },
     {
       "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": 9663.304085675061,
-      "required_pages_per_second_upper_estimate": 0.004045823627936223
+      "required_token_positions_per_second": 9641.885497409201,
+      "required_pages_per_second_upper_estimate": 0.004036856112300291
     },
     {
       "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": 19326.608171350123,
-      "required_pages_per_second_upper_estimate": 0.004025222805160463
+      "required_token_positions_per_second": 19283.770994818402,
+      "required_pages_per_second_upper_estimate": 0.0040163009509812195
     },
     {
       "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": 28989.912257025186,
-      "required_pages_per_second_upper_estimate": 0.001737900974775672
+      "required_token_positions_per_second": 28925.6564922276,
+      "required_pages_per_second_upper_estimate": 0.0017340489397889287
     },
     {
       "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": 28989.912257025186,
-      "required_pages_per_second_upper_estimate": 0.0022538381648356943
+      "required_token_positions_per_second": 28925.6564922276,
+      "required_pages_per_second_upper_estimate": 0.002248842561753922
     },
     {
       "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": 28989.912257025186,
-      "required_pages_per_second_upper_estimate": 0.37185146749047837
+      "required_token_positions_per_second": 28925.6564922276,
+      "required_pages_per_second_upper_estimate": 0.3710272635321199
     }
   ],
   "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": 0.23167637730458068,
-    "producer_p95_seconds": 0.2454739580025489,
-    "gpu_update_seconds": 0.9537110617952975,
-    "producer_batches_per_second": 4.316365835975233,
-    "gpu_batches_per_second": 1.048535599574117,
-    "producer_headroom": 4.116565844524887,
+    "producer_batch_seconds": 0.0955585915000654,
+    "producer_p95_seconds": 0.24718180399941048,
+    "gpu_update_seconds": 0.9558296458173418,
+    "producer_batches_per_second": 10.464783797062513,
+    "gpu_batches_per_second": 1.0462115340070748,
+    "producer_headroom": 10.002550590301318,
     "sustainable": true,
-    "buffer_seconds": 60.083796893103745,
+    "buffer_seconds": 60.217267686492534,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330156.15613382263,
+    "generation_bytes_per_second": 330392.75470309873,
     "upload_bytes_per_second": 4839515.31852037,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.84382546435364,
+    "estimated_full_shards_per_hour": 70.8945940095875,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -238,3 +238,4 @@
 {"step": 948800, "loss": 0.4947313070297241, "lr": 3e-05, "grad_norm": 0.3324267838111315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138878112624, "tokens_per_second": 152504.77627283018, "data_wait_s": 1.4070999895920977e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 948900, "loss": 0.5397599339485168, "lr": 3e-05, "grad_norm": 0.35230049628193194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138892548315, "tokens_per_second": 150450.3322366843, "data_wait_s": 1.689300188445486e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949000, "loss": 0.5276049375534058, "lr": 3e-05, "grad_norm": 0.35546142067112296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138907098420, "tokens_per_second": 151543.5062288765, "data_wait_s": 1.0994001058861613e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 949100, "loss": 0.5535507202148438, "lr": 3e-05, "grad_norm": 0.37402986916779357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138921774854, "tokens_per_second": 153016.35622632026, "data_wait_s": 1.4047000149730593e-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 sha256:8587ccfb817ae36246dea74c350f209e99b8d6570467a92d4a6adc24317befae
-size 511617
+oid sha256:8d1bff6f8c9cc30f6c75f519e40a899868699ce95963d555217408483d74c645
+size 513735