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": 77192.44646899514,
-      "required_pages_per_second_upper_estimate": 0.005340352557877593
+      "required_token_positions_per_second": 76720.91410903748,
+      "required_pages_per_second_upper_estimate": 0.005307730855109908
     },
     {
       "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": 9649.055808624393,
-      "required_pages_per_second_upper_estimate": 0.004039858171872967
+      "required_token_positions_per_second": 9590.114263629685,
+      "required_pages_per_second_upper_estimate": 0.004015180577823105
     },
     {
       "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": 19298.111617248785,
-      "required_pages_per_second_upper_estimate": 0.004019287724445823
+      "required_token_positions_per_second": 19180.22852725937,
+      "required_pages_per_second_upper_estimate": 0.003994735785587167
     },
     {
       "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": 28947.167425873176,
-      "required_pages_per_second_upper_estimate": 0.0017353384874156871
+      "required_token_positions_per_second": 28770.342790889055,
+      "required_pages_per_second_upper_estimate": 0.0017247381205439734
     },
     {
       "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": 28947.167425873176,
-      "required_pages_per_second_upper_estimate": 0.002250514942228268
+      "required_token_positions_per_second": 28770.342790889055,
+      "required_pages_per_second_upper_estimate": 0.0022367676046276246
     },
     {
       "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": 28947.167425873176,
-      "required_pages_per_second_upper_estimate": 0.3713031826922843
+      "required_token_positions_per_second": 28770.342790889055,
+      "required_pages_per_second_upper_estimate": 0.36903506613420883
     }
   ],
   "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": 0.09772905452331315,
-    "producer_p95_seconds": 0.34680407300038496,
-    "gpu_update_seconds": 0.9551193591151868,
-    "producer_batches_per_second": 10.232371579543432,
-    "gpu_batches_per_second": 1.046989562567751,
-    "producer_headroom": 9.773136185281974,
+    "producer_batch_seconds": 0.3851002756641151,
+    "producer_p95_seconds": 0.3818665289945784,
+    "gpu_update_seconds": 0.9609895926841554,
+    "producer_batches_per_second": 2.5967262637646127,
+    "gpu_batches_per_second": 1.0405939956195405,
+    "producer_headroom": 2.4954269145274037,
     "sustainable": true,
-    "buffer_seconds": 60.172519624256765,
+    "buffer_seconds": 60.54234433910179,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 383295.76852087333,
-    "upload_bytes_per_second": 4725714.002744023,
+    "generation_bytes_per_second": 354651.3941707648,
+    "upload_bytes_per_second": 4640816.057696938,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.24634925574922,
+    "estimated_full_shards_per_hour": 76.09993332712371,
     "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": 1.5938844269985566,
-      "bytes": 10568768,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 2.7108459590017446,
       "bytes": 1157065,
@@ -562,6 +556,12 @@
       "bytes": 5924616,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2267345689979265,
+      "bytes": 1209498,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -370,3 +370,4 @@
 {"step": 962000, "loss": 0.5000197291374207, "lr": 3e-05, "grad_norm": 0.34545883509433695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140810961962, "tokens_per_second": 152647.10812592026, "data_wait_s": 1.8989994714502245e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 962100, "loss": 0.5354152917861938, "lr": 3e-05, "grad_norm": 0.3367592078885038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140825699596, "tokens_per_second": 153290.6106381421, "data_wait_s": 1.2322998372837901e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 962200, "loss": 0.5208715200424194, "lr": 3e-05, "grad_norm": 0.33088779700782583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140840430506, "tokens_per_second": 152777.79002305097, "data_wait_s": 1.3284996384754777e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 962300, "loss": 0.5374898314476013, "lr": 3e-05, "grad_norm": 0.35127594183016225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140855028084, "tokens_per_second": 146374.7713553694, "data_wait_s": 1.471900031901896e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f8c61e8a3c4958ad7961e2e9876159bd2db627b89e135f2900ac9907ca13688
-size 793542
+oid sha256:e726d96447a10b899a49b61095d3c0062f7db8819b6ba1394cac9e4d7e4c7fc0
+size 795660