Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75969.7940652313,
-      "required_pages_per_second_upper_estimate": 0.006584869205891511
+      "required_token_positions_per_second": 75129.95868548463,
+      "required_pages_per_second_upper_estimate": 0.006512074403718379
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9496.224258153912,
-      "required_pages_per_second_upper_estimate": 0.0037881858377827957
+      "required_token_positions_per_second": 9391.244835685578,
+      "required_pages_per_second_upper_estimate": 0.0037463079765779395
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18992.448516307824,
-      "required_pages_per_second_upper_estimate": 0.004393568708570177
+      "required_token_positions_per_second": 18782.489671371157,
+      "required_pages_per_second_upper_estimate": 0.004344998424943545
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28488.67277446174,
-      "required_pages_per_second_upper_estimate": 0.0017348107034502076
+      "required_token_positions_per_second": 28173.734507056735,
+      "required_pages_per_second_upper_estimate": 0.0017156326153186315
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28488.67277446174,
-      "required_pages_per_second_upper_estimate": 0.0021931368918716772
+      "required_token_positions_per_second": 28173.734507056735,
+      "required_pages_per_second_upper_estimate": 0.002168892072241918
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28488.67277446174,
-      "required_pages_per_second_upper_estimate": 0.353733970404432
+      "required_token_positions_per_second": 28173.734507056735,
+      "required_pages_per_second_upper_estimate": 0.34982349115383904
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10268041665648298,
-    "producer_p95_seconds": 0.4178811950041563,
-    "gpu_update_seconds": 0.9704909814115543,
-    "producer_batches_per_second": 9.73895541683958,
-    "gpu_batches_per_second": 1.0304062780114922,
-    "producer_headroom": 9.451568400412018,
+    "producer_batch_seconds": 0.14860930707823172,
+    "producer_p95_seconds": 0.42082494599890197,
+    "gpu_update_seconds": 0.9813395520240651,
+    "producer_batches_per_second": 6.729053648528047,
+    "gpu_batches_per_second": 1.0190152816499107,
+    "producer_headroom": 6.603486492992413,
     "sustainable": true,
-    "buffer_seconds": 60.17044084751637,
+    "buffer_seconds": 60.84305222549204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323802.1967723479,
+    "generation_bytes_per_second": 356657.41109844355,
     "upload_bytes_per_second": 3634423.476470282,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.48041369798496,
+    "estimated_full_shards_per_hour": 76.53037786211948,
     "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-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -409,3 +409,4 @@
 {"step": 40900, "loss": 0.656970776617527, "lr": 0.0003, "grad_norm": 0.14349440390525414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5985111871, "tokens_per_second": 129470.72505672603, "data_wait_s": 1.4625002222601324e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 41000, "loss": 0.633791621774435, "lr": 0.0003, "grad_norm": 0.13616520934006598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5999586229, "tokens_per_second": 148245.87926764408, "data_wait_s": 9.536001016385853e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 41100, "loss": 0.6907564885914326, "lr": 0.0003, "grad_norm": 0.14187858018878158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6014175298, "tokens_per_second": 148623.47853386478, "data_wait_s": 8.615003025624901e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 41200, "loss": 0.6779820397496223, "lr": 0.0003, "grad_norm": 0.12701036529413634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6028842649, "tokens_per_second": 148187.00278750827, "data_wait_s": 1.006700040306896e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77f7c0edcfb205cf6a78dffb5e2a3527cea0fd73906a758c47d9413aa752c386
-size 871153
+oid sha256:40b67cb3e5351d9056be8be2cd15527ea987b48293717c52958f91430d6f8f00
+size 873271