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": 77079.94911441517,
-      "required_pages_per_second_upper_estimate": 0.005332569729857925
+      "required_token_positions_per_second": 76497.62183847235,
+      "required_pages_per_second_upper_estimate": 0.005292282977722764
     },
     {
       "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": 9634.993639301896,
-      "required_pages_per_second_upper_estimate": 0.004033970635229125
+      "required_token_positions_per_second": 9562.202729809043,
+      "required_pages_per_second_upper_estimate": 0.004003494601471508
     },
     {
       "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": 19269.987278603792,
-      "required_pages_per_second_upper_estimate": 0.004013430166394757
+      "required_token_positions_per_second": 19124.405459618087,
+      "required_pages_per_second_upper_estimate": 0.00398310931275078
     },
     {
       "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": 28904.98091790569,
-      "required_pages_per_second_upper_estimate": 0.0017328094706781042
+      "required_token_positions_per_second": 28686.608189427134,
+      "required_pages_per_second_upper_estimate": 0.0017197183590416645
     },
     {
       "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": 28904.98091790569,
-      "required_pages_per_second_upper_estimate": 0.002247235126792634
+      "required_token_positions_per_second": 28686.608189427134,
+      "required_pages_per_second_upper_estimate": 0.002230257607673548
     },
     {
       "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": 28904.98091790569,
-      "required_pages_per_second_upper_estimate": 0.37076205946442053
+      "required_token_positions_per_second": 28686.608189427134,
+      "required_pages_per_second_upper_estimate": 0.36796100857386554
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10112468608573977,
-    "producer_p95_seconds": 0.3038827409982332,
-    "gpu_update_seconds": 0.9565133455207704,
-    "producer_batches_per_second": 9.888782242074285,
-    "gpu_batches_per_second": 1.0454637195423064,
-    "producer_headroom": 9.458752185492859,
+    "producer_batch_seconds": 0.14329702025776214,
+    "producer_p95_seconds": 0.32462770599886426,
+    "gpu_update_seconds": 0.9637946674431198,
+    "producer_batches_per_second": 6.978512171440856,
+    "gpu_batches_per_second": 1.0375654003699049,
+    "producer_headroom": 6.725852817521604,
     "sustainable": true,
-    "buffer_seconds": 60.260340767808536,
+    "buffer_seconds": 60.71906404891655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268505.3762909879,
-    "upload_bytes_per_second": 5556813.051536425,
+    "generation_bytes_per_second": 300003.22315605474,
+    "upload_bytes_per_second": 5532592.155983226,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.61500326678493,
+    "estimated_full_shards_per_hour": 64.37370797167999,
     "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": 1.1054332999992766,
-      "bytes": 5601825,
-      "pages": 19,
-      "shards": 2
-    },
     {
       "seconds": 1.0161830090000876,
       "bytes": 1229232,
@@ -562,6 +556,12 @@
       "bytes": 12206708,
       "pages": 48,
       "shards": 3
+    },
+    {
+      "seconds": 0.8664777250014595,
+      "bytes": 2384518,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 677800, "loss": 0.5002837777137756, "lr": 0.0001, "grad_norm": 0.23840045262852252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99212380014, "tokens_per_second": 133157.99911329185, "data_wait_s": 1.3387998478719965e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 677900, "loss": 0.5715132355690002, "lr": 0.0001, "grad_norm": 0.2154137197133443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99227007310, "tokens_per_second": 152083.80053568436, "data_wait_s": 1.0281997674610466e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 678000, "loss": 0.5482680797576904, "lr": 0.0001, "grad_norm": 0.19308343636945574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99241725668, "tokens_per_second": 152805.12796581432, "data_wait_s": 8.801998774288222e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 678100, "loss": 0.5835674405097961, "lr": 0.0001, "grad_norm": 0.22842088470374425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99256445591, "tokens_per_second": 154111.1735606397, "data_wait_s": 1.3627999578602612e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bb785ee5f26004a4adef385f8240de4ad1840b55e22436414f7b4e67d7ef56f
-size 703803
+oid sha256:1ba9837ab2e607bccd2491785cba118bad0a9478f3a7a6d98c193ba03edf5cea
+size 705921