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": 75897.32571768628,
-      "required_pages_per_second_upper_estimate": 0.006578587833195656
+      "required_token_positions_per_second": 75286.60512399461,
+      "required_pages_per_second_upper_estimate": 0.006525652119991655
     },
     {
       "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": 9487.165714710785,
-      "required_pages_per_second_upper_estimate": 0.0037845722493660386
+      "required_token_positions_per_second": 9410.825640499326,
+      "required_pages_per_second_upper_estimate": 0.003754119052377264
     },
     {
       "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": 18974.33142942157,
-      "required_pages_per_second_upper_estimate": 0.0043893776393688815
+      "required_token_positions_per_second": 18821.651280998653,
+      "required_pages_per_second_upper_estimate": 0.004354057774110076
     },
     {
       "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": 28461.497144132354,
-      "required_pages_per_second_upper_estimate": 0.0017331558501426615
+      "required_token_positions_per_second": 28232.476921497975,
+      "required_pages_per_second_upper_estimate": 0.0017192097201604731
     },
     {
       "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": 28461.497144132354,
-      "required_pages_per_second_upper_estimate": 0.002191044836621962
+      "required_token_positions_per_second": 28232.476921497975,
+      "required_pages_per_second_upper_estimate": 0.0021734142259150152
     },
     {
       "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": 28461.497144132354,
-      "required_pages_per_second_upper_estimate": 0.35339654002672505
+      "required_token_positions_per_second": 28232.476921497975,
+      "required_pages_per_second_upper_estimate": 0.3505528753429849
     }
   ],
   "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": 0.0900087254217965,
-    "producer_p95_seconds": 0.2128496430013911,
-    "gpu_update_seconds": 0.9714176264160417,
-    "producer_batches_per_second": 11.11003400296834,
-    "gpu_batches_per_second": 1.0294233631413612,
-    "producer_headroom": 10.79248286056502,
+    "producer_batch_seconds": 0.08346005982812699,
+    "producer_p95_seconds": 0.20035949900011474,
+    "gpu_update_seconds": 0.9792977101115447,
+    "producer_batches_per_second": 11.981779093608901,
+    "gpu_batches_per_second": 1.0211399349500137,
+    "producer_headroom": 11.733728829433575,
     "sustainable": true,
-    "buffer_seconds": 60.227892837794585,
+    "buffer_seconds": 60.71645802691577,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 221639.23563130258,
-    "upload_bytes_per_second": 6553154.913139645,
+    "generation_bytes_per_second": 214664.35030467465,
+    "upload_bytes_per_second": 6554475.590911509,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.55862046913441,
+    "estimated_full_shards_per_hour": 46.06197244506053,
     "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": 0.8210835840000072,
-      "bytes": 900296,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.093376490999617,
       "bytes": 11563576,
@@ -562,6 +556,12 @@
       "bytes": 11385510,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.7792264220006473,
+      "bytes": 714539,
+      "pages": 19,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -122,3 +122,4 @@
 {"step": 12200, "loss": 0.6643102113157511, "lr": 0.0003, "grad_norm": 0.14960396821808594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1784743281, "tokens_per_second": 149334.31641902582, "data_wait_s": 1.570899985381402e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 12300, "loss": 0.7572563216090202, "lr": 0.0003, "grad_norm": 0.18024803477390786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1799280716, "tokens_per_second": 147247.47988033932, "data_wait_s": 1.1493999409140088e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 12400, "loss": 0.650935085490346, "lr": 0.0003, "grad_norm": 0.13516034694887558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1813818162, "tokens_per_second": 148622.16673962533, "data_wait_s": 8.554001396987587e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 12500, "loss": 0.6745043750852346, "lr": 0.0003, "grad_norm": 0.15191327576307453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1828371567, "tokens_per_second": 134701.158912142, "data_wait_s": 1.0660000043571927e-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 sha256:1300136d1f2fb49584edd6d1f477aa8996ac16b30a1b142d20fff13a6d038a73
-size 259964
+oid sha256:5259e190ace6e9de9c23a3375d334aa920bd4613a701089b2372f9b6e89faa0a
+size 262047