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": 75129.05850872051,
-      "required_pages_per_second_upper_estimate": 0.0051976025910661075
+      "required_token_positions_per_second": 75736.06611686354,
+      "required_pages_per_second_upper_estimate": 0.00523959678584914
     },
     {
       "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": 9391.132313590064,
-      "required_pages_per_second_upper_estimate": 0.003931870990557138
+      "required_token_positions_per_second": 9467.008264607943,
+      "required_pages_per_second_upper_estimate": 0.003963638666778291
     },
     {
       "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": 18782.264627180128,
-      "required_pages_per_second_upper_estimate": 0.0039118504002144654
+      "required_token_positions_per_second": 18934.016529215885,
+      "required_pages_per_second_upper_estimate": 0.003943456319441736
     },
     {
       "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": 28173.396940770195,
-      "required_pages_per_second_upper_estimate": 0.0016889521283128843
+      "required_token_positions_per_second": 28401.024793823828,
+      "required_pages_per_second_upper_estimate": 0.0017025980705358426
     },
     {
       "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": 28173.396940770195,
-      "required_pages_per_second_upper_estimate": 0.002190357690468187
+      "required_token_positions_per_second": 28401.024793823828,
+      "required_pages_per_second_upper_estimate": 0.0022080547548139944
     },
     {
       "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": 28173.396940770195,
-      "required_pages_per_second_upper_estimate": 0.36137808571940067
+      "required_token_positions_per_second": 28401.024793823828,
+      "required_pages_per_second_upper_estimate": 0.3642978514106262
     }
   ],
   "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": 0.7595791934371903,
-    "producer_p95_seconds": 0.3026514289995248,
-    "gpu_update_seconds": 0.9813513101783661,
-    "producer_batches_per_second": 1.3165184205150167,
-    "gpu_batches_per_second": 1.019003072221144,
-    "producer_headroom": 1.2919670768463647,
+    "producer_batch_seconds": 0.09828659907810788,
+    "producer_p95_seconds": 0.2925541320000775,
+    "gpu_update_seconds": 0.9734859992098742,
+    "producer_batches_per_second": 10.17432701283422,
+    "gpu_batches_per_second": 1.0272361398229104,
+    "producer_headroom": 9.904564898376936,
     "sustainable": true,
-    "buffer_seconds": 60.8437812310587,
+    "buffer_seconds": 60.3561319510122,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301880.7224342283,
-    "upload_bytes_per_second": 5355409.69454323,
+    "generation_bytes_per_second": 301814.4265572241,
+    "upload_bytes_per_second": 5425791.365367278,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.7765756108297,
+    "estimated_full_shards_per_hour": 64.76235005891364,
     "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": 1.360712293000688,
-      "bytes": 10168781,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.3043567909990088,
       "bytes": 7359704,
@@ -562,6 +556,12 @@
       "bytes": 1171347,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.6574927980000211,
+      "bytes": 17383397,
+      "pages": 38,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -260,3 +260,4 @@
 {"step": 746000, "loss": 0.5546980500221252, "lr": 5e-05, "grad_norm": 0.2263805816080096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109194315552, "tokens_per_second": 151491.65458015518, "data_wait_s": 1.2363001587800682e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 746100, "loss": 0.49467793107032776, "lr": 5e-05, "grad_norm": 0.2407655481504517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109208903432, "tokens_per_second": 149582.97855694933, "data_wait_s": 1.692400110187009e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 746200, "loss": 0.5053272843360901, "lr": 5e-05, "grad_norm": 0.23429421098434564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109223438106, "tokens_per_second": 149215.56514957664, "data_wait_s": 9.994000720325857e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 746300, "loss": 0.540273129940033, "lr": 5e-05, "grad_norm": 0.24720239149437304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109238001258, "tokens_per_second": 150270.01715989434, "data_wait_s": 1.0635998478392139e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33eace302f418dc47da53c209ad77638a4279ab63a43f5f1da2f95840d87df86
-size 558996
+oid sha256:b8b7e9587bae93f20b61332cb259e4861b5c521283282cf2d046cdb26b3d5a7e
+size 561114