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": 75819.97947684549,
-      "required_pages_per_second_upper_estimate": 0.0065718836570718316
+      "required_token_positions_per_second": 74107.10812776446,
+      "required_pages_per_second_upper_estimate": 0.006423416309766232
     },
     {
       "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": 9477.497434605686,
-      "required_pages_per_second_upper_estimate": 0.003780715427878445
+      "required_token_positions_per_second": 9263.388515970557,
+      "required_pages_per_second_upper_estimate": 0.0036953041790212847
     },
     {
       "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": 18954.994869211372,
-      "required_pages_per_second_upper_estimate": 0.004384904466476089
+      "required_token_positions_per_second": 18526.777031941114,
+      "required_pages_per_second_upper_estimate": 0.004285843806200151
     },
     {
       "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": 28432.492303817056,
-      "required_pages_per_second_upper_estimate": 0.001731389607544096
+      "required_token_positions_per_second": 27790.165547911667,
+      "required_pages_per_second_upper_estimate": 0.0016922752781374966
     },
     {
       "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": 28432.492303817056,
-      "required_pages_per_second_upper_estimate": 0.002188811963723953
+      "required_token_positions_per_second": 27790.165547911667,
+      "required_pages_per_second_upper_estimate": 0.0021393638719800945
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28432.492303817056,
-      "required_pages_per_second_upper_estimate": 0.3530363969829651
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27790.165547911667,
+      "required_pages_per_second_upper_estimate": 0.3450608484039841
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11362801548415291,
-    "producer_p95_seconds": 0.38528945100551937,
-    "gpu_update_seconds": 0.972408598745607,
-    "producer_batches_per_second": 8.800646528403593,
-    "gpu_batches_per_second": 1.0283742876091237,
-    "producer_headroom": 8.557824358740328,
+    "producer_batch_seconds": 0.49584234303915764,
+    "producer_p95_seconds": 0.38086970699805534,
+    "gpu_update_seconds": 0.994884321661684,
+    "producer_batches_per_second": 2.01677007629223,
+    "gpu_batches_per_second": 1.005141983069722,
+    "producer_headroom": 2.0064529292995776,
     "sustainable": true,
-    "buffer_seconds": 60.28933312222764,
+    "buffer_seconds": 60.68794362136272,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382704.94645894266,
+    "generation_bytes_per_second": 338293.2283208696,
     "upload_bytes_per_second": 6025842.797389354,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.1195725948926,
+    "estimated_full_shards_per_hour": 72.58985173434797,
     "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-v1/1790485060735632541-7579de36/train.jsonl

@@ -326,3 +326,4 @@
 {"step": 117600, "loss": 0.6011027507483959, "lr": 0.0003, "grad_norm": 0.12590950636835682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17209104158, "tokens_per_second": 150209.93611194243, "data_wait_s": 1.4432000170927495e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 117700, "loss": 0.6190260946750641, "lr": 0.0003, "grad_norm": 0.15998024372133723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17223795313, "tokens_per_second": 150029.1593504651, "data_wait_s": 1.867800165200606e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 117800, "loss": 0.5871414011344314, "lr": 0.0003, "grad_norm": 0.13882546866497125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17238468392, "tokens_per_second": 149223.67364011874, "data_wait_s": 1.260400313185528e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 117900, "loss": 0.623930312693119, "lr": 0.0003, "grad_norm": 0.144380699745971, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17253039451, "tokens_per_second": 148450.04961165026, "data_wait_s": 2.0261999452486634e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58d74851ec87b3fefb79af0d62f6ae33a1678442c90dae28aec5d9b19d857f4a
-size 699567
+oid sha256:f09258ee4f5353b3320a00ca7846628e06de39c8ad0256ede9905bbddbe87924
+size 701685