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": 75544.35413990366,
-      "required_pages_per_second_upper_estimate": 0.005226333706449982
+      "required_token_positions_per_second": 75635.94824387149,
+      "required_pages_per_second_upper_estimate": 0.005232670399090079
     },
     {
       "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": 9443.044267487958,
-      "required_pages_per_second_upper_estimate": 0.0039536054416093175
+      "required_token_positions_per_second": 9454.493530483936,
+      "required_pages_per_second_upper_estimate": 0.003958399008937935
     },
     {
       "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": 18886.088534975916,
-      "required_pages_per_second_upper_estimate": 0.0039334741821877555
+      "required_token_positions_per_second": 18908.987060967873,
+      "required_pages_per_second_upper_estimate": 0.00393824334127714
     },
     {
       "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": 28329.132802463875,
-      "required_pages_per_second_upper_estimate": 0.0016982882554265283
+      "required_token_positions_per_second": 28363.480591451807,
+      "required_pages_per_second_upper_estimate": 0.0017003473529303178
     },
     {
       "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": 28329.132802463875,
-      "required_pages_per_second_upper_estimate": 0.002202465468705209
+      "required_token_positions_per_second": 28363.480591451807,
+      "required_pages_per_second_upper_estimate": 0.0022051358582190635
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28329.132802463875,
-      "required_pages_per_second_upper_estimate": 0.36337569813706694
+      "required_token_positions_per_second": 28363.480591451807,
+      "required_pages_per_second_upper_estimate": 0.3638162746944217
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.13759020112490816,
-    "producer_p95_seconds": 0.36903263100248296,
-    "gpu_update_seconds": 0.9759564541839899,
-    "producer_batches_per_second": 7.267959431879691,
-    "gpu_batches_per_second": 1.0246358797187454,
-    "producer_headroom": 7.0932119162903895,
+    "producer_batch_seconds": 0.7463992443042571,
+    "producer_p95_seconds": 0.3818901979975635,
+    "gpu_update_seconds": 0.9747745842000985,
+    "producer_batches_per_second": 1.339765557951673,
+    "gpu_batches_per_second": 1.0258782042625798,
+    "producer_headroom": 1.305969414677955,
     "sustainable": true,
-    "buffer_seconds": 60.509300159407374,
+    "buffer_seconds": 60.43602422040611,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356785.76220977836,
-    "upload_bytes_per_second": 4475359.081231629,
+    "generation_bytes_per_second": 309085.17503924435,
+    "upload_bytes_per_second": 4541513.915282942,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.55791902275098,
+    "estimated_full_shards_per_hour": 66.32248342879294,
     "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": 1.2824578229992767,
-      "bytes": 5780729,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.419093739001255,
       "bytes": 6340288,
@@ -562,6 +556,12 @@
       "bytes": 6634160,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4960427599944524,
+      "bytes": 13030938,
+      "pages": 29,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -416,3 +416,4 @@
 {"step": 646000, "loss": 0.6285852789878845, "lr": 0.0001, "grad_norm": 0.19419523950274928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94557920764, "tokens_per_second": 152505.10645218327, "data_wait_s": 1.3965996913611889e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 646100, "loss": 0.5348060727119446, "lr": 0.0001, "grad_norm": 0.17619847701323677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94572621539, "tokens_per_second": 152832.1986084975, "data_wait_s": 1.0864001524168998e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 646200, "loss": 0.53627610206604, "lr": 0.0001, "grad_norm": 0.19759993762466208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94587097212, "tokens_per_second": 150246.5098884262, "data_wait_s": 9.414994565304369e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 646300, "loss": 0.5539202690124512, "lr": 0.0001, "grad_norm": 0.1678955062400825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94601667259, "tokens_per_second": 150495.375290938, "data_wait_s": 1.1004995030816644e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42d4cc69885095c5b77f64a76e8f0b280568463cf837f4c4fd49074340e05b85
-size 892536
+oid sha256:1eeabf4c380baedb21307a36f30bb233ae626b7052fc73a548a4b85e5a79f797
+size 894654