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": 76950.49855190274,
-      "required_pages_per_second_upper_estimate": 0.0053236140395766275
+      "required_token_positions_per_second": 77045.88326776368,
+      "required_pages_per_second_upper_estimate": 0.005330212975542921
     },
     {
       "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": 9618.812318987842,
-      "required_pages_per_second_upper_estimate": 0.004027195854318023
+      "required_token_positions_per_second": 9630.73540847046,
+      "required_pages_per_second_upper_estimate": 0.004032187802901974
     },
     {
       "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": 19237.624637975685,
-      "required_pages_per_second_upper_estimate": 0.004006689881812148
+      "required_token_positions_per_second": 19261.47081694092,
+      "required_pages_per_second_upper_estimate": 0.004011656412024578
     },
     {
       "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": 28856.436956963527,
-      "required_pages_per_second_upper_estimate": 0.001729899334341958
+      "required_token_positions_per_second": 28892.20622541138,
+      "required_pages_per_second_upper_estimate": 0.0017320436473688963
     },
     {
       "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": 28856.436956963527,
-      "required_pages_per_second_upper_estimate": 0.0022434610473516994
+      "required_token_positions_per_second": 28892.20622541138,
+      "required_pages_per_second_upper_estimate": 0.002246241950641135
     },
     {
       "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": 28856.436956963527,
-      "required_pages_per_second_upper_estimate": 0.3701393896558988
+      "required_token_positions_per_second": 28892.20622541138,
+      "required_pages_per_second_upper_estimate": 0.3705981994254997
     }
   ],
   "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": 0.09446278102370798,
-    "producer_p95_seconds": 0.3386847720030346,
-    "gpu_update_seconds": 0.9581224473844159,
-    "producer_batches_per_second": 10.586179965938364,
-    "gpu_batches_per_second": 1.0437079339179516,
-    "producer_headroom": 10.142856657416738,
+    "producer_batch_seconds": 0.10667915717215237,
+    "producer_p95_seconds": 0.34440630100289127,
+    "gpu_update_seconds": 0.9569362680127531,
+    "producer_batches_per_second": 9.373902330201771,
+    "gpu_batches_per_second": 1.0450016719260482,
+    "producer_headroom": 8.970227112579332,
     "sustainable": true,
-    "buffer_seconds": 60.3617141852182,
+    "buffer_seconds": 60.286984884803445,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342887.02533316903,
-    "upload_bytes_per_second": 4839646.050524728,
+    "generation_bytes_per_second": 342666.3443740242,
+    "upload_bytes_per_second": 4729783.59042181,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.57557363506606,
+    "estimated_full_shards_per_hour": 73.52822063842339,
     "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": 1.3592490819974046,
-      "bytes": 11983959,
-      "pages": 49,
-      "shards": 4
-    },
     {
       "seconds": 1.0659817249979824,
       "bytes": 6806858,
@@ -562,6 +556,12 @@
       "bytes": 10804637,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1380872770023416,
+      "bytes": 998627,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 961600, "loss": 0.48917630314826965, "lr": 3e-05, "grad_norm": 0.34890945974183674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140752217671, "tokens_per_second": 152169.75341136171, "data_wait_s": 8.320996130350977e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 961700, "loss": 0.5648232102394104, "lr": 3e-05, "grad_norm": 0.35461705620083384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140766952259, "tokens_per_second": 138410.45664885247, "data_wait_s": 1.4290002582129091e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 961800, "loss": 0.5268715023994446, "lr": 3e-05, "grad_norm": 0.3619495455912913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140781581022, "tokens_per_second": 152507.38925987462, "data_wait_s": 1.0173003829549998e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 961900, "loss": 0.5561473965644836, "lr": 3e-05, "grad_norm": 0.3620200640664721, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140796255527, "tokens_per_second": 153307.72205360627, "data_wait_s": 1.733899989631027e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3566cf9310ce1047a435aae35fdecce49496bef88278b383f7dd5e36f922934
-size 785070
+oid sha256:5d42e1d2d0f79ecd7868b4aef050fa4550b6f566c55ce78c4934d3a947bbcae3
+size 787188