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": 76213.92194697386,
-      "required_pages_per_second_upper_estimate": 0.006606029591944496
+      "required_token_positions_per_second": 75895.57656905257,
+      "required_pages_per_second_upper_estimate": 0.006578436221425263
     },
     {
       "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": 9526.740243371733,
-      "required_pages_per_second_upper_estimate": 0.003800359120540822
+      "required_token_positions_per_second": 9486.947071131572,
+      "required_pages_per_second_upper_estimate": 0.0037844850291732773
     },
     {
       "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": 19053.480486743465,
-      "required_pages_per_second_upper_estimate": 0.004407687380804469
+      "required_token_positions_per_second": 18973.894142263143,
+      "required_pages_per_second_upper_estimate": 0.0043892764806807705
     },
     {
       "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": 28580.220730115198,
-      "required_pages_per_second_upper_estimate": 0.0017403854936345008
+      "required_token_positions_per_second": 28460.841213394717,
+      "required_pages_per_second_upper_estimate": 0.0017331159074021418
     },
     {
       "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": 28580.220730115198,
-      "required_pages_per_second_upper_estimate": 0.0022001845069188415
+      "required_token_positions_per_second": 28460.841213394717,
+      "required_pages_per_second_upper_estimate": 0.002190994341258046
     },
     {
       "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": 28580.220730115198,
-      "required_pages_per_second_upper_estimate": 0.35487068962234997
+      "required_token_positions_per_second": 28460.841213394717,
+      "required_pages_per_second_upper_estimate": 0.35338839556222257
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0749976534218959,
-    "producer_p95_seconds": 0.18347764000009192,
-    "gpu_update_seconds": 0.9673823117421585,
-    "producer_batches_per_second": 13.333750515826745,
-    "gpu_batches_per_second": 1.0337174743241897,
-    "producer_headroom": 12.898834398193676,
+    "producer_batch_seconds": 0.5166383134453412,
+    "producer_p95_seconds": 0.1841241980000632,
+    "gpu_update_seconds": 0.9714400144640783,
+    "producer_batches_per_second": 1.935590090737235,
+    "gpu_batches_per_second": 1.0293996387946585,
+    "producer_headroom": 1.8803096657423064,
     "sustainable": true,
-    "buffer_seconds": 60.94508563975599,
+    "buffer_seconds": 60.229280896772856,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432830.50353887305,
+    "generation_bytes_per_second": 394137.7785946498,
     "upload_bytes_per_second": 4508842.591861955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.87535028099674,
+    "estimated_full_shards_per_hour": 84.57279222850438,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -47,3 +47,4 @@
 {"step": 159700, "loss": 0.6141259353607893, "lr": 0.0003, "grad_norm": 0.13582600490727087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23373370513, "tokens_per_second": 151190.07967858604, "data_wait_s": 1.5458000234502833e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159800, "loss": 0.5677892863750458, "lr": 0.0003, "grad_norm": 0.1604483381015062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23388092979, "tokens_per_second": 150897.233162046, "data_wait_s": 1.2524999874585774e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 159900, "loss": 0.6478084838017821, "lr": 0.0003, "grad_norm": 0.1428728914111215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23402615397, "tokens_per_second": 147312.37066334608, "data_wait_s": 1.5530999917245936e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 160000, "loss": 0.5920981746166945, "lr": 0.0003, "grad_norm": 0.1573822250209816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23417085332, "tokens_per_second": 148750.36305334992, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6702cc2784157db84b0eb8fcdfce90729164444907497674a0d5a9eb84452269
-size 103947
+oid sha256:f583fea85a504945d8ba10a27dead149c37816013e9f056ab2616d60a8d9367c
+size 106065