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": 75429.95227981002,
-      "required_pages_per_second_upper_estimate": 0.00653807708814769
+      "required_token_positions_per_second": 75020.85254225972,
+      "required_pages_per_second_upper_estimate": 0.006502617359750635
     },
     {
       "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": 9428.744034976253,
-      "required_pages_per_second_upper_estimate": 0.0037612669678379817
+      "required_token_positions_per_second": 9377.606567782464,
+      "required_pages_per_second_upper_estimate": 0.0037408674676011108
     },
     {
       "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": 18857.488069952506,
-      "required_pages_per_second_upper_estimate": 0.004362347984528606
+      "required_token_positions_per_second": 18755.21313556493,
+      "required_pages_per_second_upper_estimate": 0.004338688478435409
     },
     {
       "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": 28286.232104928757,
-      "required_pages_per_second_upper_estimate": 0.0017224831287997568
+      "required_token_positions_per_second": 28132.819703347395,
+      "required_pages_per_second_upper_estimate": 0.0017131411184361177
     },
     {
       "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": 28286.232104928757,
-      "required_pages_per_second_upper_estimate": 0.0021775524487393882
+      "required_token_positions_per_second": 28132.819703347395,
+      "required_pages_per_second_upper_estimate": 0.0021657423373929474
     },
     {
       "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": 28286.232104928757,
-      "required_pages_per_second_upper_estimate": 0.35122033481416937
+      "required_token_positions_per_second": 28132.819703347395,
+      "required_pages_per_second_upper_estimate": 0.3493154662248084
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.08391777881264773,
-    "producer_p95_seconds": 0.21857567699953506,
-    "gpu_update_seconds": 0.9774366517759873,
-    "producer_batches_per_second": 11.916425984445674,
-    "gpu_batches_per_second": 1.0230842051840552,
-    "producer_headroom": 11.647551515372953,
+    "producer_batch_seconds": 0.42875664313292816,
+    "producer_p95_seconds": 0.20970142899932398,
+    "gpu_update_seconds": 0.9827667575287624,
+    "producer_batches_per_second": 2.332325378548055,
+    "gpu_batches_per_second": 1.0175354348722294,
+    "producer_headroom": 2.2921318497777152,
     "sustainable": true,
-    "buffer_seconds": 60.60107241011121,
+    "buffer_seconds": 60.93153896678327,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369261.25810425216,
-    "upload_bytes_per_second": 5750901.706386707,
+    "generation_bytes_per_second": 334472.47165540396,
+    "upload_bytes_per_second": 5713807.3478162605,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.23487002702402,
+    "estimated_full_shards_per_hour": 71.77000629660215,
     "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": [
@@ -532,6 +532,12 @@
       "bytes": 6233203,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8259697120010969,
+      "bytes": 2261871,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -96,3 +96,4 @@
 {"step": 364600, "loss": 0.585701584815979, "lr": 0.0003, "grad_norm": 0.14848347318623806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53367190260, "tokens_per_second": 150535.86390228407, "data_wait_s": 1.3957998817204498e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 364700, "loss": 0.5882318466901779, "lr": 0.0003, "grad_norm": 0.16263330731698614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53381922131, "tokens_per_second": 150098.51633814754, "data_wait_s": 1.0455998562974855e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 364800, "loss": 0.5859208181500435, "lr": 0.0003, "grad_norm": 0.15828551047307055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53396639105, "tokens_per_second": 150056.78764171584, "data_wait_s": 7.472999641322531e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 364900, "loss": 0.5903571639209986, "lr": 0.0003, "grad_norm": 0.1696520185735211, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53411391267, "tokens_per_second": 150548.5994235989, "data_wait_s": 1.2202001016703434e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb32e8fba59af8e529112d65d4022267cbace7c5246b9aad8f3829b3890b7640
-size 208512
+oid sha256:e239c274d7b10003b6b297f06004a1b08e051f763157aeecd8641fb572203305
+size 210630