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": 77460.6781496336,
-      "required_pages_per_second_upper_estimate": 0.005358909447927409
+      "required_token_positions_per_second": 77423.23466546951,
+      "required_pages_per_second_upper_estimate": 0.005356319020812084
     },
     {
       "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": 9682.5847687042,
-      "required_pages_per_second_upper_estimate": 0.004053896047294077
+      "required_token_positions_per_second": 9677.90433318369,
+      "required_pages_per_second_upper_estimate": 0.0040519364466802465
     },
     {
       "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": 19365.1695374084,
-      "required_pages_per_second_upper_estimate": 0.004033254120778793
+      "required_token_positions_per_second": 19355.80866636738,
+      "required_pages_per_second_upper_estimate": 0.004031304498203719
     },
     {
       "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": 29047.7543061126,
-      "required_pages_per_second_upper_estimate": 0.0017413685172987675
+      "required_token_positions_per_second": 29033.71299955107,
+      "required_pages_per_second_upper_estimate": 0.0017405267624102374
     },
     {
       "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": 29047.7543061126,
-      "required_pages_per_second_upper_estimate": 0.002258335129731955
+      "required_token_positions_per_second": 29033.71299955107,
+      "required_pages_per_second_upper_estimate": 0.0022572434799079766
     },
     {
       "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": 29047.7543061126,
-      "required_pages_per_second_upper_estimate": 0.3725934031902182
+      "required_token_positions_per_second": 29033.71299955107,
+      "required_pages_per_second_upper_estimate": 0.3724132963860272
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08053056315614526,
-    "producer_p95_seconds": 0.21704928699909942,
-    "gpu_update_seconds": 0.9518119613873887,
-    "producer_batches_per_second": 12.417645683925535,
-    "gpu_batches_per_second": 1.0506276875764107,
-    "producer_headroom": 11.819263694230806,
+    "producer_batch_seconds": 0.0835661673515915,
+    "producer_p95_seconds": 0.2227969209998264,
+    "gpu_update_seconds": 0.9522722774185824,
+    "producer_batches_per_second": 11.966565318146724,
+    "gpu_batches_per_second": 1.0501198278194108,
+    "producer_headroom": 11.395428408389805,
     "sustainable": true,
-    "buffer_seconds": 60.915965528792874,
+    "buffer_seconds": 60.94542575478928,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300227.39289391396,
+    "generation_bytes_per_second": 310912.3313573506,
     "upload_bytes_per_second": 5898566.347581735,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.42180957902016,
+    "estimated_full_shards_per_hour": 66.71454864063632,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -142,3 +142,4 @@
 {"step": 499200, "loss": 0.5943893194198608, "lr": 0.0003, "grad_norm": 0.16097518033957775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73066090337, "tokens_per_second": 130870.35168583132, "data_wait_s": 1.1677999282255769e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 499300, "loss": 0.5743064284324646, "lr": 0.0003, "grad_norm": 0.15934545158083602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73080747017, "tokens_per_second": 152847.28738402095, "data_wait_s": 1.1795000318670645e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 499400, "loss": 0.5696477890014648, "lr": 0.0003, "grad_norm": 0.1744138905646286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73095317497, "tokens_per_second": 151790.53945064975, "data_wait_s": 9.437000699108467e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 499500, "loss": 0.5315882563591003, "lr": 0.0003, "grad_norm": 0.13697746125448093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73109944416, "tokens_per_second": 152287.63724088288, "data_wait_s": 1.1536001693457365e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68fbcbaa8bcb0dce2eb10c208f7568637a28911e4052ab91cee7bc4f9ee90d3e
-size 306723
+oid sha256:9f30616a11f6529a6d1ce98d73e7778e5fba01ada2e5f9bf6c1538d55b8344ea
+size 308841