Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77505.40364377553,
-      "required_pages_per_second_upper_estimate": 0.005362003661389591
+      "required_token_positions_per_second": 76974.06547394946,
+      "required_pages_per_second_upper_estimate": 0.005325244453926613
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9688.175455471941,
-      "required_pages_per_second_upper_estimate": 0.004056236751096914
+      "required_token_positions_per_second": 9621.758184243683,
+      "required_pages_per_second_upper_estimate": 0.004028429226583982
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19376.350910943882,
-      "required_pages_per_second_upper_estimate": 0.004035582906013585
+      "required_token_positions_per_second": 19243.516368487366,
+      "required_pages_per_second_upper_estimate": 0.00400791697390235
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29064.526366415823,
-      "required_pages_per_second_upper_estimate": 0.0017423739767044924
+      "required_token_positions_per_second": 28865.274552731047,
+      "required_pages_per_second_upper_estimate": 0.0017304291347139936
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29064.526366415823,
-      "required_pages_per_second_upper_estimate": 0.0022596390836480337
+      "required_token_positions_per_second": 28865.274552731047,
+      "required_pages_per_second_upper_estimate": 0.0022441481315501484
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29064.526366415823,
-      "required_pages_per_second_upper_estimate": 0.37280853717135265
+      "required_token_positions_per_second": 28865.274552731047,
+      "required_pages_per_second_upper_estimate": 0.37025274884533355
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05031710174220905,
-    "producer_p95_seconds": 0.18368494200058194,
-    "gpu_update_seconds": 0.9512627060025783,
-    "producer_batches_per_second": 19.87395866167584,
-    "gpu_batches_per_second": 1.051234315914924,
-    "producer_headroom": 18.90535569548914,
+    "producer_batch_seconds": 0.10695051328126937,
+    "producer_p95_seconds": 0.22286923100000422,
+    "gpu_update_seconds": 0.9578291018674591,
+    "producer_batches_per_second": 9.350118754176505,
+    "gpu_batches_per_second": 1.0440275807556079,
+    "producer_headroom": 8.955815848666967,
     "sustainable": true,
-    "buffer_seconds": 60.880813184165014,
+    "buffer_seconds": 60.34323341764992,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 408065.7785464495,
-    "upload_bytes_per_second": 5471496.780001619,
+    "generation_bytes_per_second": 380384.973611145,
+    "upload_bytes_per_second": 5644585.712937581,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.56141679091562,
+    "estimated_full_shards_per_hour": 81.62176042795909,
     "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": 1.386034201999962,
-      "bytes": 47132,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 1.0035069749999934,
       "bytes": 155991,
@@ -562,6 +556,12 @@
       "bytes": 877584,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.2262302330000239,
+      "bytes": 12002858,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 611900, "loss": 0.5776656866073608, "lr": 0.0001, "grad_norm": 0.14340920056441128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89564452425, "tokens_per_second": 150589.9909592441, "data_wait_s": 9.762000445334706e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 612000, "loss": 0.55414217710495, "lr": 0.0001, "grad_norm": 0.15120684381030386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89578955010, "tokens_per_second": 151040.63235765524, "data_wait_s": 1.0994000149366912e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 612100, "loss": 0.5246615409851074, "lr": 0.0001, "grad_norm": 0.1585942307912722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89593479598, "tokens_per_second": 150958.17484313183, "data_wait_s": 1.0028999895439483e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 612200, "loss": 0.5195173025131226, "lr": 0.0001, "grad_norm": 0.15260576830122513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89608129676, "tokens_per_second": 152321.39499359208, "data_wait_s": 7.714000275882427e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5cd8073f622eb52e2c245529cd7b521a536aa3070e8a0a5d2ad29d3aed18f0ca
-size 164817
+oid sha256:4c514f73cafc0667f14c8ad945eee44859b11e98394270ac92c4512e8b6c7308
+size 166935