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": 77284.70485310226,
-      "required_pages_per_second_upper_estimate": 0.005346735206959068
+      "required_token_positions_per_second": 76810.31082412277,
+      "required_pages_per_second_upper_estimate": 0.005313915527288466
     },
     {
       "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": 9660.588106637782,
-      "required_pages_per_second_upper_estimate": 0.004044686504229405
+      "required_token_positions_per_second": 9601.288853015347,
+      "required_pages_per_second_upper_estimate": 0.00401985914504692
     },
     {
       "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": 19321.176213275565,
-      "required_pages_per_second_upper_estimate": 0.0040240914715439
+      "required_token_positions_per_second": 19202.577706030694,
+      "required_pages_per_second_upper_estimate": 0.003999390530137755
     },
     {
       "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": 28981.76431991335,
-      "required_pages_per_second_upper_estimate": 0.001737412518386997
+      "required_token_positions_per_second": 28803.866559046037,
+      "required_pages_per_second_upper_estimate": 0.0017267478192571544
     },
     {
       "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": 28981.76431991335,
-      "required_pages_per_second_upper_estimate": 0.002253204698564237
+      "required_token_positions_per_second": 28803.866559046037,
+      "required_pages_per_second_upper_estimate": 0.0022393739301463573
     },
     {
       "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": 28981.76431991335,
-      "required_pages_per_second_upper_estimate": 0.371746954501781
+      "required_token_positions_per_second": 28803.866559046037,
+      "required_pages_per_second_upper_estimate": 0.3694650730371088
     }
   ],
   "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": 0.20962030816386346,
-    "producer_p95_seconds": 0.2620828690014605,
-    "gpu_update_seconds": 0.9539791882512508,
-    "producer_batches_per_second": 4.770530149293953,
-    "gpu_batches_per_second": 1.0482408969876067,
-    "producer_headroom": 4.550986479351564,
+    "producer_batch_seconds": 0.09557471592199818,
+    "producer_p95_seconds": 0.27043781399697764,
+    "gpu_update_seconds": 0.9598711319997061,
+    "producer_batches_per_second": 10.463018282116941,
+    "gpu_batches_per_second": 1.0418065161691998,
+    "producer_headroom": 10.043149202589207,
     "sustainable": true,
-    "buffer_seconds": 60.1006888598288,
+    "buffer_seconds": 60.471881315981484,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 543515.9470282523,
-    "upload_bytes_per_second": 5412187.102245186,
+    "generation_bytes_per_second": 542995.8515206317,
+    "upload_bytes_per_second": 5396120.741991421,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 116.62586982856442,
+    "estimated_full_shards_per_hour": 116.51426943983282,
     "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": 1.141444616998342,
-      "bytes": 5373163,
-      "pages": 23,
-      "shards": 2
-    },
     {
       "seconds": 1.302073504000873,
       "bytes": 12612142,
@@ -562,6 +556,12 @@
       "bytes": 5751342,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.5010855370019272,
+      "bytes": 6022741,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -262,3 +262,4 @@
 {"step": 671200, "loss": 0.47457993030548096, "lr": 0.0001, "grad_norm": 0.19315317112429023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98246591406, "tokens_per_second": 151463.09626353803, "data_wait_s": 1.1727999662980437e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 671300, "loss": 0.546379804611206, "lr": 0.0001, "grad_norm": 0.1983805729814073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98261127449, "tokens_per_second": 151619.72613656745, "data_wait_s": 1.3402997865341604e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 671400, "loss": 0.5036755204200745, "lr": 0.0001, "grad_norm": 0.1960686886120762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98275792779, "tokens_per_second": 152326.90409161427, "data_wait_s": 2.0974999642930925e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 671500, "loss": 0.5429437160491943, "lr": 0.0001, "grad_norm": 0.18978454644553686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98290413269, "tokens_per_second": 151672.6207776673, "data_wait_s": 1.601499752723612e-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 sha256:685e06e4c97f934f648bcd8c1565e8ba817518b2f1cf505ad139e967fdeea782
-size 563232
+oid sha256:b9cca6ad9bfacd7e60c69b02d71d42ab79866e8bee07274af98fced1f6fb5ddc
+size 565350