Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75123.44943856973,
-      "required_pages_per_second_upper_estimate": 0.006511510198693332
+      "required_token_positions_per_second": 75699.72273013931,
+      "required_pages_per_second_upper_estimate": 0.006561460107055279
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9390.431179821217,
-      "required_pages_per_second_upper_estimate": 0.003745983397088406
+      "required_token_positions_per_second": 9462.465341267414,
+      "required_pages_per_second_upper_estimate": 0.0037747189010959845
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18780.862359642433,
-      "required_pages_per_second_upper_estimate": 0.004344621975014813
+      "required_token_positions_per_second": 18924.93068253483,
+      "required_pages_per_second_upper_estimate": 0.004377949646000084
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28171.293539463648,
-      "required_pages_per_second_upper_estimate": 0.0017154839731989796
+      "required_token_positions_per_second": 28387.396023802245,
+      "required_pages_per_second_upper_estimate": 0.0017286434806930897
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28171.293539463648,
-      "required_pages_per_second_upper_estimate": 0.002168704159799571
+      "required_token_positions_per_second": 28387.396023802245,
+      "required_pages_per_second_upper_estimate": 0.00218534032725392
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28171.293539463648,
-      "required_pages_per_second_upper_estimate": 0.3497931825057259
+      "required_token_positions_per_second": 28387.396023802245,
+      "required_pages_per_second_upper_estimate": 0.35247645211272144
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.42641095915598726,
-    "producer_p95_seconds": 0.36865752899757354,
-    "gpu_update_seconds": 0.9814245824839177,
-    "producer_batches_per_second": 2.345155485636066,
-    "gpu_batches_per_second": 1.0189269943382395,
-    "producer_headroom": 2.3015932433502453,
+    "producer_batch_seconds": 0.09355845628874704,
+    "producer_p95_seconds": 0.35424404199875426,
+    "gpu_update_seconds": 0.9739533691930645,
+    "producer_batches_per_second": 10.688504702490237,
+    "gpu_batches_per_second": 1.0267432010923845,
+    "producer_headroom": 10.41010516662628,
     "sustainable": true,
-    "buffer_seconds": 60.8483241140029,
+    "buffer_seconds": 60.38510888997,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359935.18582775997,
-    "upload_bytes_per_second": 5075397.911749826,
+    "generation_bytes_per_second": 360055.7300867683,
+    "upload_bytes_per_second": 5061438.807425536,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.23371201634025,
+    "estimated_full_shards_per_hour": 77.2595780081967,
     "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": 1.198589940999227,
-      "bytes": 12036533,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 0.8354435640030715,
       "bytes": 5772648,
@@ -562,6 +556,12 @@
       "bytes": 2483287,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1772192700009327,
+      "bytes": 10757429,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -381,3 +381,4 @@
 {"step": 193100, "loss": 0.5864177253097296, "lr": 0.0003, "grad_norm": 0.1442607025269688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28262060521, "tokens_per_second": 150170.15379304672, "data_wait_s": 1.2289005098864436e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 193200, "loss": 0.6590218525379896, "lr": 0.0003, "grad_norm": 0.19263882744823843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28276651113, "tokens_per_second": 149004.7270510985, "data_wait_s": 1.0096999176312238e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 193300, "loss": 0.6072239205241203, "lr": 0.0003, "grad_norm": 0.16099361596421796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28291318837, "tokens_per_second": 150242.16291948615, "data_wait_s": 1.2869000784121454e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 193400, "loss": 0.5866144802421331, "lr": 0.0003, "grad_norm": 0.14337903761812446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28305891675, "tokens_per_second": 149803.2884866767, "data_wait_s": 9.187999239657074e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e61ccd8f81e13f5e62b134da942a498adb6ef2d7b7ca87b1201eeb345c72cf61
-size 816840
+oid sha256:200214fd8e6073066caea31ce4fe05d4a557654e1519e660a235b6dab5d23328
+size 818958