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": 76045.39641777654,
-      "required_pages_per_second_upper_estimate": 0.005260996973811485
+      "required_token_positions_per_second": 75838.97141279065,
+      "required_pages_per_second_upper_estimate": 0.00524671601299456
     },
     {
       "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": 9505.674552222068,
-      "required_pages_per_second_upper_estimate": 0.003979827433958422
+      "required_token_positions_per_second": 9479.871426598831,
+      "required_pages_per_second_upper_estimate": 0.003969024204090508
     },
     {
       "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": 19011.349104444136,
-      "required_pages_per_second_upper_estimate": 0.0039595626554646286
+      "required_token_positions_per_second": 18959.742853197662,
+      "required_pages_per_second_upper_estimate": 0.003948814434278353
     },
     {
       "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": 28517.023656666202,
-      "required_pages_per_second_upper_estimate": 0.001709552025243235
+      "required_token_positions_per_second": 28439.614279796493,
+      "required_pages_per_second_upper_estimate": 0.0017049114512971706
     },
     {
       "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": 28517.023656666202,
-      "required_pages_per_second_upper_estimate": 0.0022170731561749137
+      "required_token_positions_per_second": 28439.614279796493,
+      "required_pages_per_second_upper_estimate": 0.0022110549176111586
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28517.023656666202,
-      "required_pages_per_second_upper_estimate": 0.36578576027329307
+      "required_token_positions_per_second": 28439.614279796493,
+      "required_pages_per_second_upper_estimate": 0.3647928359025217
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07112346886717091,
-    "producer_p95_seconds": 0.2632646069978364,
-    "gpu_update_seconds": 0.969526144553902,
-    "producer_batches_per_second": 14.060056630077822,
-    "gpu_batches_per_second": 1.0314317005449292,
-    "producer_headroom": 13.631592496768878,
+    "producer_batch_seconds": 0.12564865240602785,
+    "producer_p95_seconds": 0.2857425729998795,
+    "gpu_update_seconds": 0.9721650838155405,
+    "producer_batches_per_second": 7.958700557874238,
+    "gpu_batches_per_second": 1.0286318822264358,
+    "producer_headroom": 7.737170794908597,
     "sustainable": true,
-    "buffer_seconds": 60.11062096234192,
+    "buffer_seconds": 60.27423519656351,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 313238.6679981962,
-    "upload_bytes_per_second": 5338642.320568301,
+    "generation_bytes_per_second": 326950.939380991,
+    "upload_bytes_per_second": 5293873.765262302,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.21372632941642,
+    "estimated_full_shards_per_hour": 70.15606056282327,
     "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": 1.154910495999502,
-      "bytes": 10565027,
-      "pages": 23,
-      "shards": 3
-    },
     {
       "seconds": 1.3083442860006471,
       "bytes": 8430845,
@@ -562,6 +556,12 @@
       "bytes": 5982392,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0109681320027448,
+      "bytes": 6177508,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -217,3 +217,4 @@
 {"step": 626100, "loss": 0.5501269698143005, "lr": 0.0001, "grad_norm": 0.15863774938725253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91644575305, "tokens_per_second": 151858.25455336555, "data_wait_s": 2.2718002583133057e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626200, "loss": 0.5858222246170044, "lr": 0.0001, "grad_norm": 0.16567491129721723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91659315953, "tokens_per_second": 152543.8206018042, "data_wait_s": 1.3728000340051949e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 626300, "loss": 0.5707839727401733, "lr": 0.0001, "grad_norm": 0.16334502314435753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91674039090, "tokens_per_second": 151943.28471567607, "data_wait_s": 1.7668997315922752e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 626400, "loss": 0.5396880507469177, "lr": 0.0001, "grad_norm": 0.14936795053782706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91688532080, "tokens_per_second": 150298.5900370969, "data_wait_s": 1.7931997717823833e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:edb045488278c3db550741a4c726965ef7d67dce79e3fbacb20c0eea2e93fdb7
-size 467922
+oid sha256:2a0f8510542f25b2f68ae03eef5449400fbdaf016cebc7401bee2caad9859b8d
+size 470040