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": 77186.16917553553,
-      "required_pages_per_second_upper_estimate": 0.00533991828015073
+      "required_token_positions_per_second": 76979.6203619136,
+      "required_pages_per_second_upper_estimate": 0.005325628753965088
     },
     {
       "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": 9648.271146941941,
-      "required_pages_per_second_upper_estimate": 0.004039529650412123
+      "required_token_positions_per_second": 9622.4525452392,
+      "required_pages_per_second_upper_estimate": 0.004028719941032898
     },
     {
       "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": 19296.542293883882,
-      "required_pages_per_second_upper_estimate": 0.004018960875774753
+      "required_token_positions_per_second": 19244.9050904784,
+      "required_pages_per_second_upper_estimate": 0.004008206208070021
     },
     {
       "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": 28944.813440825827,
-      "required_pages_per_second_upper_estimate": 0.0017351973696064255
+      "required_token_positions_per_second": 28867.3576357176,
+      "required_pages_per_second_upper_estimate": 0.0017305540123583537
     },
     {
       "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": 28944.813440825827,
-      "required_pages_per_second_upper_estimate": 0.0022503319302449223
+      "required_token_positions_per_second": 28867.3576357176,
+      "required_pages_per_second_upper_estimate": 0.002244310082089839
     },
     {
       "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": 28944.813440825827,
-      "required_pages_per_second_upper_estimate": 0.37127298829960914
+      "required_token_positions_per_second": 28867.3576357176,
+      "required_pages_per_second_upper_estimate": 0.37027946839724474
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11691697604697993,
-    "producer_p95_seconds": 0.2487919979976141,
-    "gpu_update_seconds": 0.9551970357840791,
-    "producer_batches_per_second": 8.553077866110538,
-    "gpu_batches_per_second": 1.0469044213261656,
-    "producer_headroom": 8.169874624539203,
+    "producer_batch_seconds": 0.06926807051559081,
+    "producer_p95_seconds": 0.25783929500175873,
+    "gpu_update_seconds": 0.957759984439695,
+    "producer_batches_per_second": 14.436666021683406,
+    "gpu_batches_per_second": 1.044102923745573,
+    "producer_headroom": 13.826861024288574,
     "sustainable": true,
-    "buffer_seconds": 60.17741325439698,
+    "buffer_seconds": 60.338879019700784,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342385.8448016709,
-    "upload_bytes_per_second": 4945757.993722686,
+    "generation_bytes_per_second": 378208.54085940286,
+    "upload_bytes_per_second": 4980356.22537974,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.46803195989222,
+    "estimated_full_shards_per_hour": 81.15474862419667,
     "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": 1.0568985930003691,
-      "bytes": 2271584,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.734822353000709,
       "bytes": 6223398,
@@ -562,6 +556,12 @@
       "bytes": 8432247,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2434194350025791,
+      "bytes": 6179718,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 419600, "loss": 0.5711472630500793, "lr": 0.0003, "grad_norm": 0.16745750840028553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61418193013, "tokens_per_second": 152023.9459269514, "data_wait_s": 1.3665001461049542e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 419700, "loss": 0.5668135285377502, "lr": 0.0003, "grad_norm": 0.17400848672498184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61432850501, "tokens_per_second": 152177.04341231627, "data_wait_s": 1.072399754775688e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 419800, "loss": 0.5982234477996826, "lr": 0.0003, "grad_norm": 0.15461834384812445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61447664225, "tokens_per_second": 152977.95596793166, "data_wait_s": 1.2906999472761527e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 419900, "loss": 0.6178654432296753, "lr": 0.0003, "grad_norm": 0.1400032057203984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61462422501, "tokens_per_second": 153448.85455168787, "data_wait_s": 1.3737000699620694e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26bf45a7d1c121b45a55801033db432cacf2e35f2c7061ab312d45e353b430d1
-size 484083
+oid sha256:bfd015653c3f3006834fcfcae5e4ef614f140fa0aced7e4613215f79972b3a94
+size 486201