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": 75650.13489469071,
-      "required_pages_per_second_upper_estimate": 0.006557161959157813
+      "required_token_positions_per_second": 75363.84613289469,
+      "required_pages_per_second_upper_estimate": 0.006532347174877567
     },
     {
       "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": 9456.266861836339,
-      "required_pages_per_second_upper_estimate": 0.0037722462349754023
+      "required_token_positions_per_second": 9420.480766611836,
+      "required_pages_per_second_upper_estimate": 0.003757970626540544
     },
     {
       "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": 18912.533723672677,
-      "required_pages_per_second_upper_estimate": 0.00437508182774635
+      "required_token_positions_per_second": 18840.96153322367,
+      "required_pages_per_second_upper_estimate": 0.0043585248611135025
     },
     {
       "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": 28368.800585509016,
-      "required_pages_per_second_upper_estimate": 0.0017275111160637568
+      "required_token_positions_per_second": 28261.442299835508,
+      "required_pages_per_second_upper_estimate": 0.0017209735597316336
     },
     {
       "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": 28368.800585509016,
-      "required_pages_per_second_upper_estimate": 0.0021839087989386397
+      "required_token_positions_per_second": 28261.442299835508,
+      "required_pages_per_second_upper_estimate": 0.0021756440609207377
     },
     {
       "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": 28368.800585509016,
-      "required_pages_per_second_upper_estimate": 0.352245559004048
+      "required_token_positions_per_second": 28261.442299835508,
+      "required_pages_per_second_upper_estimate": 0.3509125284010518
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14594370521842848,
-    "producer_p95_seconds": 0.4139527570005157,
-    "gpu_update_seconds": 0.9745917849668553,
-    "producer_batches_per_second": 6.851957050859696,
-    "gpu_batches_per_second": 1.0260706230291166,
-    "producer_headroom": 6.67786105271358,
+    "producer_batch_seconds": 0.43232177324227905,
+    "producer_p95_seconds": 0.4090578160030418,
+    "gpu_update_seconds": 0.9782940200529299,
+    "producer_batches_per_second": 2.3130919187814913,
+    "gpu_batches_per_second": 1.0221875831827079,
+    "producer_headroom": 2.26288399197669,
     "sustainable": true,
-    "buffer_seconds": 60.42469066794503,
+    "buffer_seconds": 60.654229243281655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398438.4646999911,
-    "upload_bytes_per_second": 3714811.316808087,
+    "generation_bytes_per_second": 358046.27287202416,
+    "upload_bytes_per_second": 3675236.32781576,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.49561935186195,
+    "estimated_full_shards_per_hour": 76.8283952676825,
     "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": 1.378931018000003,
-      "bytes": 13331524,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.0631361519990605,
       "bytes": 9774660,
@@ -562,6 +556,12 @@
       "bytes": 5952538,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9529956519982079,
+      "bytes": 7196338,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -391,3 +391,4 @@
 {"step": 39100, "loss": 0.6157486774027348, "lr": 0.0003, "grad_norm": 0.12539534924991447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5721435182, "tokens_per_second": 148431.78037223127, "data_wait_s": 7.2879993240348995e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 39200, "loss": 0.6467400174587965, "lr": 0.0003, "grad_norm": 0.13735031642942835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5736088739, "tokens_per_second": 149885.70432057107, "data_wait_s": 1.4270997780840844e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 39300, "loss": 0.6332474946975708, "lr": 0.0003, "grad_norm": 0.13531782894139807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5750821886, "tokens_per_second": 149299.85389619958, "data_wait_s": 1.409299875376746e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 39400, "loss": 0.6049629338085651, "lr": 0.0003, "grad_norm": 0.12598470520687913, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5765473879, "tokens_per_second": 134883.06262163044, "data_wait_s": 1.0157003998756409e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d880349fbe41e28b1b0afa08a259608ddf69a22f9e929f6e8f2eb6fb5c82382
-size 832246
+oid sha256:6333c0206bdf6aa543b05e61f1a28433d21b9142c6a4fbe5673e8fd3a1330f17
+size 834364