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": 75550.19877153687,
-      "required_pages_per_second_upper_estimate": 0.006548499749288627
+      "required_token_positions_per_second": 73779.41610322442,
+      "required_pages_per_second_upper_estimate": 0.006395012795606944
     },
     {
       "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": 9443.77484644211,
-      "required_pages_per_second_upper_estimate": 0.00376726298326237
+      "required_token_positions_per_second": 9222.427012903052,
+      "required_pages_per_second_upper_estimate": 0.0036789640230186103
     },
     {
       "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": 18887.54969288422,
-      "required_pages_per_second_upper_estimate": 0.004369302211927364
+      "required_token_positions_per_second": 18444.854025806104,
+      "required_pages_per_second_upper_estimate": 0.004266892360526481
     },
     {
       "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": 28331.324539326328,
-      "required_pages_per_second_upper_estimate": 0.0017252290214728498
+      "required_token_positions_per_second": 27667.281038709156,
+      "required_pages_per_second_upper_estimate": 0.0016847922562522565
     },
     {
       "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": 28331.324539326328,
-      "required_pages_per_second_upper_estimate": 0.002181023789163161
+      "required_token_positions_per_second": 27667.281038709156,
+      "required_pages_per_second_upper_estimate": 0.002129903882295592
     },
     {
       "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": 28331.324539326328,
-      "required_pages_per_second_upper_estimate": 0.3517802319347173
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27667.281038709156,
+      "required_pages_per_second_upper_estimate": 0.3435350340676851
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 690670,
-    "producer_batch_seconds": 0.08801387023427765,
-    "producer_p95_seconds": 0.3242520589992637,
-    "gpu_update_seconds": 0.975880953310961,
-    "producer_batches_per_second": 11.361845551595147,
-    "gpu_batches_per_second": 1.0247151526087357,
-    "producer_headroom": 11.087808668262573,
+    "producer_batch_seconds": 0.4320460435155269,
+    "producer_p95_seconds": 0.3225658029987244,
+    "gpu_update_seconds": 0.9993031104616974,
+    "producer_batches_per_second": 2.314568123024744,
+    "gpu_batches_per_second": 1.0006973755320152,
+    "producer_headroom": 2.3129551247141196,
     "sustainable": true,
-    "buffer_seconds": 60.50461910527958,
+    "buffer_seconds": 60.95748973816354,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335339.4595371224,
-    "upload_bytes_per_second": 3979676.12432033,
+    "generation_bytes_per_second": 296585.41787994764,
+    "upload_bytes_per_second": 3972390.366839625,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.95604171357397,
+    "estimated_full_shards_per_hour": 63.640326521862235,
     "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": 0.8681581979999464,
-      "bytes": 818733,
-      "pages": 23,
-      "shards": 1
-    },
     {
       "seconds": 1.2214710799999011,
       "bytes": 12703928,
@@ -562,6 +556,12 @@
       "bytes": 6068321,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 2.3319496820004133,
+      "bytes": 5855101,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 294700, "loss": 0.6058078296482563, "lr": 0.0003, "grad_norm": 0.16915365988143374, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43136199010, "tokens_per_second": 148780.71310533409, "data_wait_s": 1.2674001482082531e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 294800, "loss": 0.6485474109649658, "lr": 0.0003, "grad_norm": 0.17541547180358677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43150702227, "tokens_per_second": 147358.1234483947, "data_wait_s": 9.000999853014946e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 294900, "loss": 0.6201039385050535, "lr": 0.0003, "grad_norm": 0.1613886835568409, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43165218484, "tokens_per_second": 148639.20341720953, "data_wait_s": 1.2045999028487131e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 295000, "loss": 0.5687651764601469, "lr": 0.0003, "grad_norm": 0.15474079343900016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43179922574, "tokens_per_second": 147643.66191736262, "data_wait_s": 1.5697001799708232e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8e8c784a5785f2dde788b4ff0c5442a1cb7d6288c7ceadf7990610ff50f38f8d
-size 313077
+oid sha256:4d6c55b1c0224608f6d29333a43970fc7d15ae23e4a7a6f4cbe42bb712d6c81c
+size 315195