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": 76924.34362659873,
-      "required_pages_per_second_upper_estimate": 0.005321804581156308
+      "required_token_positions_per_second": 77280.82338233547,
+      "required_pages_per_second_upper_estimate": 0.005346466677805168
     },
     {
       "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": 9615.54295332484,
-      "required_pages_per_second_upper_estimate": 0.0040258270391870426
+      "required_token_positions_per_second": 9660.102922791933,
+      "required_pages_per_second_upper_estimate": 0.004044483367884939
     },
     {
       "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": 19231.08590664968,
-      "required_pages_per_second_upper_estimate": 0.004005328036514883
+      "required_token_positions_per_second": 19320.205845583867,
+      "required_pages_per_second_upper_estimate": 0.004023889369544036
     },
     {
       "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": 28846.62885997452,
-      "required_pages_per_second_upper_estimate": 0.001729311353903564
+      "required_token_positions_per_second": 28980.308768375802,
+      "required_pages_per_second_upper_estimate": 0.0017373252602948196
     },
     {
       "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": 28846.62885997452,
-      "required_pages_per_second_upper_estimate": 0.002242698510952063
+      "required_token_positions_per_second": 28980.308768375802,
+      "required_pages_per_second_upper_estimate": 0.002253091535834487
     },
     {
       "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": 28846.62885997452,
-      "required_pages_per_second_upper_estimate": 0.3700135819188379
+      "required_token_positions_per_second": 28980.308768375802,
+      "required_pages_per_second_upper_estimate": 0.3717282842495068
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11056786793758988,
-    "producer_p95_seconds": 0.21660160400097084,
-    "gpu_update_seconds": 0.9584482170934833,
-    "producer_batches_per_second": 9.044218891554015,
-    "gpu_batches_per_second": 1.0433531850395876,
-    "producer_headroom": 8.668415471613146,
+    "producer_batch_seconds": 0.07966740895315638,
+    "producer_p95_seconds": 0.1932289079995826,
+    "gpu_update_seconds": 0.9540271023671888,
+    "producer_batches_per_second": 12.552184301462468,
+    "gpu_batches_per_second": 1.0481882511709997,
+    "producer_headroom": 11.975124017503154,
     "sustainable": true,
-    "buffer_seconds": 60.382237676889446,
+    "buffer_seconds": 60.10370744913289,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 453880.16331378784,
+    "generation_bytes_per_second": 439126.55146092444,
     "upload_bytes_per_second": 5746819.356311452,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.39211725769259,
+    "estimated_full_shards_per_hour": 94.22633560057449,
     "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": [

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 539100, "loss": 0.5736814737319946, "lr": 0.0003, "grad_norm": 0.15854465597854384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78906393519, "tokens_per_second": 152540.7430293124, "data_wait_s": 1.4277999071055092e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 539200, "loss": 0.6041315197944641, "lr": 0.0003, "grad_norm": 0.17044956637918734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78920870866, "tokens_per_second": 151258.39625752752, "data_wait_s": 1.7632000890444033e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 539300, "loss": 0.5678359866142273, "lr": 0.0003, "grad_norm": 0.14804303306266192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78935470031, "tokens_per_second": 153069.068175832, "data_wait_s": 1.471900031901896e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 539400, "loss": 0.5947290062904358, "lr": 0.0003, "grad_norm": 0.16231946269614997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78950087711, "tokens_per_second": 153394.0111428723, "data_wait_s": 7.07400067767594e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b7ea143d044a1ba9fe65b5b8b8898b00fdb824b93dea0cdfdf76c5fce6374c40
-size 304605
+oid sha256:f67bf0ceff5ad36362b24b9d26690dc3f21be1016122672a19473913e8037b74
+size 306723