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": 76331.69043662318,
-      "required_pages_per_second_upper_estimate": 0.0052808034583289405
+      "required_token_positions_per_second": 76845.8529088853,
+      "required_pages_per_second_upper_estimate": 0.005316374411181336
     },
     {
       "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": 9541.461304577897,
-      "required_pages_per_second_upper_estimate": 0.003994810599857439
+      "required_token_positions_per_second": 9605.731613610662,
+      "required_pages_per_second_upper_estimate": 0.004021719236132788
     },
     {
       "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": 19082.922609155794,
-      "required_pages_per_second_upper_estimate": 0.003974469528975818
+      "required_token_positions_per_second": 19211.463227221324,
+      "required_pages_per_second_upper_estimate": 0.004001241149874811
     },
     {
       "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": 28624.383913733694,
-      "required_pages_per_second_upper_estimate": 0.0017159881087248105
+      "required_token_positions_per_second": 28817.194840831988,
+      "required_pages_per_second_upper_estimate": 0.0017275468293991142
     },
     {
       "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": 28624.383913733694,
-      "required_pages_per_second_upper_estimate": 0.0022254199439340466
+      "required_token_positions_per_second": 28817.194840831988,
+      "required_pages_per_second_upper_estimate": 0.0022404101454302993
     },
     {
       "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": 28624.383913733694,
-      "required_pages_per_second_upper_estimate": 0.36716286237649204
+      "required_token_positions_per_second": 28817.194840831988,
+      "required_pages_per_second_upper_estimate": 0.36963603392506494
     }
   ],
   "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": 0.12658614054720374,
-    "producer_p95_seconds": 0.3546257509951829,
-    "gpu_update_seconds": 0.9658897841547872,
-    "producer_batches_per_second": 7.89975897580274,
-    "gpu_batches_per_second": 1.0353148116946504,
-    "producer_headroom": 7.630296492012952,
+    "producer_batch_seconds": 0.13933187675030467,
+    "producer_p95_seconds": 0.3537698090003687,
+    "gpu_update_seconds": 0.9594271806367732,
+    "producer_batches_per_second": 7.177108521922018,
+    "gpu_batches_per_second": 1.0422885865462959,
+    "producer_headroom": 6.885912994311801,
     "sustainable": true,
-    "buffer_seconds": 60.85105640175159,
+    "buffer_seconds": 60.443912380116714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310945.4211490102,
-    "upload_bytes_per_second": 4542792.843566985,
+    "generation_bytes_per_second": 311132.3455350127,
+    "upload_bytes_per_second": 4591038.245352895,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.72164893963556,
+    "estimated_full_shards_per_hour": 66.7617585614947,
     "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": 1.3758785059981165,
-      "bytes": 2329519,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.3759045859987964,
       "bytes": 6265583,
@@ -562,6 +556,12 @@
       "bytes": 5750188,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2483798539979034,
+      "bytes": 6326945,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -419,3 +419,4 @@
 {"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}
 {"step": 646400, "loss": 0.5078825950622559, "lr": 0.0001, "grad_norm": 0.1642803529331283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94616241350, "tokens_per_second": 151187.51690315478, "data_wait_s": 8.134993549901992e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 646500, "loss": 0.588966965675354, "lr": 0.0001, "grad_norm": 0.19958182031668154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94630940027, "tokens_per_second": 152533.12140559233, "data_wait_s": 1.2881006114184856e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 646600, "loss": 0.5488482117652893, "lr": 0.0001, "grad_norm": 0.1857746478974556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94645761373, "tokens_per_second": 153547.3558024513, "data_wait_s": 1.0648996976669878e-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 sha256:6f302bde2b97fd1039b300410a210316d808cd7560b5cb5cab0174e771d04540
-size 898890
+oid sha256:44b691d231de7f89108093786b725a03bfbae51053dac7500d90cdf6f0a90b86
+size 901008