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": 75463.01470765423,
-      "required_pages_per_second_upper_estimate": 0.006540942855597269
+      "required_token_positions_per_second": 76007.12130079906,
+      "required_pages_per_second_upper_estimate": 0.00658810463606551
     },
     {
       "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": 9432.87683845678,
-      "required_pages_per_second_upper_estimate": 0.0037629156049372825
+      "required_token_positions_per_second": 9500.890162599882,
+      "required_pages_per_second_upper_estimate": 0.0037900471368277813
     },
     {
       "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": 18865.75367691356,
-      "required_pages_per_second_upper_estimate": 0.004364260087229331
+      "required_token_positions_per_second": 19001.780325199765,
+      "required_pages_per_second_upper_estimate": 0.0043957274583231
     },
     {
       "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": 28298.63051537034,
-      "required_pages_per_second_upper_estimate": 0.001723238126946217
+      "required_token_positions_per_second": 28502.670487799645,
+      "required_pages_per_second_upper_estimate": 0.00173566309075216
     },
     {
       "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": 28298.63051537034,
-      "required_pages_per_second_upper_estimate": 0.0021785069126961784
+      "required_token_positions_per_second": 28502.670487799645,
+      "required_pages_per_second_upper_estimate": 0.00219421447459258
     },
     {
       "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": 28298.63051537034,
-      "required_pages_per_second_upper_estimate": 0.35137428157704337
+      "required_token_positions_per_second": 28502.670487799645,
+      "required_pages_per_second_upper_estimate": 0.35390777515675587
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.276708702195279,
-    "producer_p95_seconds": 0.2183840820016485,
-    "gpu_update_seconds": 0.9770084098233323,
-    "producer_batches_per_second": 3.613908749766314,
-    "gpu_batches_per_second": 1.0235326430617164,
-    "producer_headroom": 3.5308192408558132,
+    "producer_batch_seconds": 0.09706910465622798,
+    "producer_p95_seconds": 0.21767691900095087,
+    "gpu_update_seconds": 0.9700143715247496,
+    "producer_batches_per_second": 10.301939051993097,
+    "gpu_batches_per_second": 1.0309125610459942,
+    "producer_headroom": 9.99302893500536,
     "sustainable": true,
-    "buffer_seconds": 60.574521409046596,
+    "buffer_seconds": 60.14089103453448,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 230518.03101274202,
-    "upload_bytes_per_second": 6343139.739296014,
+    "generation_bytes_per_second": 231687.3749603033,
+    "upload_bytes_per_second": 6344006.355549283,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 49.463803270213084,
+    "estimated_full_shards_per_hour": 49.7147172604258,
     "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": 0.7844010799999523,
-      "bytes": 865744,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.143092911000167,
       "bytes": 10241728,
@@ -562,6 +556,12 @@
       "bytes": 9654277,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 0.8135462400005053,
+      "bytes": 1110572,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -146,3 +146,4 @@
 {"step": 14600, "loss": 0.6604617889970541, "lr": 0.0003, "grad_norm": 0.1363267703414383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2136234563, "tokens_per_second": 148322.30530545712, "data_wait_s": 8.441998943453655e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 14700, "loss": 0.7486466094851494, "lr": 0.0003, "grad_norm": 0.15834357247497818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2150977681, "tokens_per_second": 151136.25996471135, "data_wait_s": 8.709999747225083e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 14800, "loss": 0.6797532290220261, "lr": 0.0003, "grad_norm": 0.135252270166779, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2165650987, "tokens_per_second": 149397.83953967682, "data_wait_s": 1.0089001079904847e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 14900, "loss": 0.7111192606389523, "lr": 0.0003, "grad_norm": 0.1485962875691798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2180136936, "tokens_per_second": 148151.89915795487, "data_wait_s": 1.0784999176394194e-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:e656f0dfff8ba9b8fffcd8975c0ecdacba17d7ffe4b0fa37f1b3c1c479a0fcbd
-size 309956
+oid sha256:37d7031da53e6de9c27907329f609f832035402df9f09dd44ba377f74b45ba2c
+size 312039