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": 77133.08036731233,
-      "required_pages_per_second_upper_estimate": 0.005336245473214842
+      "required_token_positions_per_second": 76934.690366754,
+      "required_pages_per_second_upper_estimate": 0.005322520392647995
     },
     {
       "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": 9641.635045914041,
-      "required_pages_per_second_upper_estimate": 0.004036751253489289
+      "required_token_positions_per_second": 9616.83629584425,
+      "required_pages_per_second_upper_estimate": 0.0040263685346918566
     },
     {
       "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": 19283.270091828083,
-      "required_pages_per_second_upper_estimate": 0.004016196626098036
+      "required_token_positions_per_second": 19233.6725916885,
+      "required_pages_per_second_upper_estimate": 0.004005866774792997
     },
     {
       "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": 28924.905137742127,
-      "required_pages_per_second_upper_estimate": 0.0017340038972347765
+      "required_token_positions_per_second": 28850.50888753275,
+      "required_pages_per_second_upper_estimate": 0.001729543955631221
     },
     {
       "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": 28924.905137742127,
-      "required_pages_per_second_upper_estimate": 0.0022487841472475355
+      "required_token_positions_per_second": 28850.50888753275,
+      "required_pages_per_second_upper_estimate": 0.002243000166028278
     },
     {
       "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": 28924.905137742127,
-      "required_pages_per_second_upper_estimate": 0.37101762596352184
+      "required_token_positions_per_second": 28850.50888753275,
+      "required_pages_per_second_upper_estimate": 0.3700633507463058
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06917693182813878,
-    "producer_p95_seconds": 0.16739374099961424,
-    "gpu_update_seconds": 0.9558544744862109,
-    "producer_batches_per_second": 14.455685928430187,
-    "gpu_batches_per_second": 1.0461843582806034,
-    "producer_headroom": 13.817532076457349,
+    "producer_batch_seconds": 0.2615879527343665,
+    "producer_p95_seconds": 0.16687891600031435,
+    "gpu_update_seconds": 0.9583193179634903,
+    "producer_batches_per_second": 3.8228060181940613,
+    "gpu_batches_per_second": 1.043493521684489,
+    "producer_headroom": 3.663468856062459,
     "sustainable": true,
-    "buffer_seconds": 60.218831892631286,
+    "buffer_seconds": 60.37411703169989,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 486702.53127662704,
-    "upload_bytes_per_second": 4704068.82613897,
+    "generation_bytes_per_second": 440424.77772319113,
+    "upload_bytes_per_second": 4697119.616504709,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.43503335689648,
+    "estimated_full_shards_per_hour": 94.50490473529625,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 298657,
       "pages": 9,
       "shards": 1
+    },
+    {
+      "seconds": 1.3505803890002426,
+      "bytes": 6226670,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 765000, "loss": 0.5579110383987427, "lr": 3e-05, "grad_norm": 0.2431269280429593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111973474399, "tokens_per_second": 153445.67202687074, "data_wait_s": 2.4105999727908056e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765100, "loss": 0.5478135943412781, "lr": 3e-05, "grad_norm": 0.24365099445058058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111988119969, "tokens_per_second": 141452.28978523877, "data_wait_s": 1.1377000191714615e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 765200, "loss": 0.5334945321083069, "lr": 3e-05, "grad_norm": 0.25597445530657653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112002882488, "tokens_per_second": 152509.53592797738, "data_wait_s": 1.3847000445821323e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 765300, "loss": 0.532582700252533, "lr": 3e-05, "grad_norm": 0.23488419984191805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112017439376, "tokens_per_second": 150846.18801539313, "data_wait_s": 1.6494999727001414e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b72bc7e1ee2586d3c85ae3d27dd921628b3528a10f6c61309c4fbee12b30239d
-size 111084
+oid sha256:472ec722a049c80ae00b8821b83af0a5fb8893787a8d31966fbfcb72c98787e6
+size 113202