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": 77341.28166839482,
-      "required_pages_per_second_upper_estimate": 0.005350649322317305
+      "required_token_positions_per_second": 77348.55352139378,
+      "required_pages_per_second_upper_estimate": 0.005351152405980796
     },
     {
       "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": 9667.660208549352,
-      "required_pages_per_second_upper_estimate": 0.004047647445617498
+      "required_token_positions_per_second": 9668.569190174223,
+      "required_pages_per_second_upper_estimate": 0.004048028017242137
     },
     {
       "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": 19335.320417098705,
-      "required_pages_per_second_upper_estimate": 0.0040270373361925215
+      "required_token_positions_per_second": 19337.138380348446,
+      "required_pages_per_second_upper_estimate": 0.004027415969994528
     },
     {
       "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": 29002.98062564806,
-      "required_pages_per_second_upper_estimate": 0.0017386844035203698
+      "required_token_positions_per_second": 29005.70757052267,
+      "required_pages_per_second_upper_estimate": 0.001738847879701799
     },
     {
       "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": 29002.98062564806,
-      "required_pages_per_second_upper_estimate": 0.002254854173014444
+      "required_token_positions_per_second": 29005.70757052267,
+      "required_pages_per_second_upper_estimate": 0.0022550661809838824
     },
     {
       "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": 29002.98062564806,
-      "required_pages_per_second_upper_estimate": 0.3720190944914516
+      "required_token_positions_per_second": 29005.70757052267,
+      "required_pages_per_second_upper_estimate": 0.3720540728123378
     }
   ],
   "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": 0.09527713949219674,
-    "producer_p95_seconds": 0.2191641060016991,
-    "gpu_update_seconds": 0.9532813319038729,
-    "producer_batches_per_second": 10.495697135007928,
-    "gpu_batches_per_second": 1.0490082691568308,
-    "producer_headroom": 10.00535214412002,
+    "producer_batch_seconds": 0.0831242215391228,
+    "producer_p95_seconds": 0.2285701950022485,
+    "gpu_update_seconds": 0.9531917100377012,
+    "producer_batches_per_second": 12.03018784999202,
+    "gpu_batches_per_second": 1.0491068999755018,
+    "producer_headroom": 11.467075328808669,
     "sustainable": true,
-    "buffer_seconds": 60.05672390994399,
+    "buffer_seconds": 60.05107773237518,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 402564.9410295652,
-    "upload_bytes_per_second": 5735685.118812759,
+    "generation_bytes_per_second": 426643.7952058317,
+    "upload_bytes_per_second": 5746159.7853829805,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.38106511273591,
+    "estimated_full_shards_per_hour": 91.54782669192518,
     "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": 1.1886329499993735,
-      "bytes": 6263136,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9854559350005729,
       "bytes": 2421861,
@@ -562,6 +556,12 @@
       "bytes": 8617085,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.9656906599993818,
+      "bytes": 5770022,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -198,3 +198,4 @@
 {"step": 704800, "loss": 0.5306462049484253, "lr": 8e-05, "grad_norm": 0.20430684528132304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103163135910, "tokens_per_second": 152653.186544431, "data_wait_s": 1.72220024978742e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 704900, "loss": 0.4992697238922119, "lr": 8e-05, "grad_norm": 0.2101155164146234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103177784453, "tokens_per_second": 152551.52564595608, "data_wait_s": 1.8695998733164743e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 705000, "loss": 0.5618807673454285, "lr": 8e-05, "grad_norm": 0.20961666563614817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103192342154, "tokens_per_second": 151723.58814000429, "data_wait_s": 1.3707998732570559e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 705100, "loss": 0.575244128704071, "lr": 8e-05, "grad_norm": 0.22217285645204765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 103206859900, "tokens_per_second": 142009.79687112584, "data_wait_s": 1.3758999557467178e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26d0194f1c998ed0a57d4929d3f6e658a7d1d99c1e238f08898af17a7aecf024
-size 426897
+oid sha256:4badc8bbc312d280427eaacce680d55e87830ddb6777ab325678cb2233aa0ad7
+size 429015