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": 76898.48561001035,
-      "required_pages_per_second_upper_estimate": 0.00532001566356986
+      "required_token_positions_per_second": 76574.32850851708,
+      "required_pages_per_second_upper_estimate": 0.005297589723140454
     },
     {
       "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": 9612.310701251294,
-      "required_pages_per_second_upper_estimate": 0.004024473762740947
+      "required_token_positions_per_second": 9571.791063564635,
+      "required_pages_per_second_upper_estimate": 0.004007509036587796
     },
     {
       "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": 19224.621402502587,
-      "required_pages_per_second_upper_estimate": 0.004003981650781482
+      "required_token_positions_per_second": 19143.58212712927,
+      "required_pages_per_second_upper_estimate": 0.003987103306870628
     },
     {
       "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": 28836.932103753883,
-      "required_pages_per_second_upper_estimate": 0.0017287300481742263
+      "required_token_positions_per_second": 28715.373190693903,
+      "required_pages_per_second_upper_estimate": 0.0017214427769459935
     },
     {
       "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": 28836.932103753883,
-      "required_pages_per_second_upper_estimate": 0.002241944630807432
+      "required_token_positions_per_second": 28715.373190693903,
+      "required_pages_per_second_upper_estimate": 0.0022324939599981703
     },
     {
       "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": 28836.932103753883,
-      "required_pages_per_second_upper_estimate": 0.3698892023416052
+      "required_token_positions_per_second": 28715.373190693903,
+      "required_pages_per_second_upper_estimate": 0.36832997512466364
     }
   ],
   "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": 0.13460957599198764,
-    "producer_p95_seconds": 0.33911576499667717,
-    "gpu_update_seconds": 0.9587705065339072,
-    "producer_batches_per_second": 7.42889198358015,
-    "gpu_batches_per_second": 1.043002463243413,
-    "producer_headroom": 7.122602530082823,
+    "producer_batch_seconds": 0.20510517542186335,
+    "producer_p95_seconds": 0.3381489559978945,
+    "gpu_update_seconds": 0.9628292070729619,
+    "producer_batches_per_second": 4.875547376818675,
+    "gpu_batches_per_second": 1.0386058011680377,
+    "producer_headroom": 4.694319414868985,
     "sustainable": true,
-    "buffer_seconds": 60.40254191163615,
+    "buffer_seconds": 60.6582400455966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441473.410928687,
-    "upload_bytes_per_second": 4584945.522488508,
+    "generation_bytes_per_second": 449546.75942362566,
+    "upload_bytes_per_second": 4568176.753714023,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.72991701026399,
+    "estimated_full_shards_per_hour": 96.4622696593435,
     "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": 1.564457227003004,
-      "bytes": 6307452,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.4970643039996503,
       "bytes": 11855130,
@@ -562,6 +556,12 @@
       "bytes": 6275099,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.0290950830021757,
+      "bytes": 2268611,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -425,3 +425,4 @@
 {"step": 802500, "loss": 0.5279629230499268, "lr": 3e-05, "grad_norm": 0.2662826119642307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117462572395, "tokens_per_second": 150028.95271998906, "data_wait_s": 1.4810997527092695e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802600, "loss": 0.5096213817596436, "lr": 3e-05, "grad_norm": 0.2796951484113033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117477393048, "tokens_per_second": 153937.12177577338, "data_wait_s": 9.019000572152436e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 802700, "loss": 0.5637000203132629, "lr": 3e-05, "grad_norm": 0.2688420309560021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117492060387, "tokens_per_second": 152672.75408613237, "data_wait_s": 8.587005140725523e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 802800, "loss": 0.545868456363678, "lr": 3e-05, "grad_norm": 0.2683358329979781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117506774323, "tokens_per_second": 152528.6755177442, "data_wait_s": 1.4285993529483676e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:abad22fd5be53162fe871be12d05e2eae5961815061209bb7a4d64650c0320a3
-size 910815
+oid sha256:7cedf7774fedee040586074ad06de7649f4528facd3adcdcd050aa8019def060
+size 912933