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": 75667.09066889404,
-      "required_pages_per_second_upper_estimate": 0.005234824904313521
+      "required_token_positions_per_second": 76521.52373802113,
+      "required_pages_per_second_upper_estimate": 0.005293936566593597
     },
     {
       "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": 9458.386333611756,
-      "required_pages_per_second_upper_estimate": 0.00396002884431658
+      "required_token_positions_per_second": 9565.190467252642,
+      "required_pages_per_second_upper_estimate": 0.00400474550474809
     },
     {
       "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": 18916.77266722351,
-      "required_pages_per_second_upper_estimate": 0.003939864877739946
+      "required_token_positions_per_second": 19130.380934505283,
+      "required_pages_per_second_upper_estimate": 0.003984353846585913
     },
     {
       "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": 28375.159000835265,
-      "required_pages_per_second_upper_estimate": 0.0017010474557409574
+      "required_token_positions_per_second": 28695.571401757923,
+      "required_pages_per_second_upper_estimate": 0.0017202556899348627
     },
     {
       "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": 28375.159000835265,
-      "required_pages_per_second_upper_estimate": 0.0022060438031807333
+      "required_token_positions_per_second": 28695.571401757923,
+      "required_pages_per_second_upper_estimate": 0.002230954458007262
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28375.159000835265,
-      "required_pages_per_second_upper_estimate": 0.3639660727906936
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28695.571401757923,
+      "required_pages_per_second_upper_estimate": 0.36807597903769734
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11767518306260172,
-    "producer_p95_seconds": 0.32287089599412866,
-    "gpu_update_seconds": 0.9743733946719432,
-    "producer_batches_per_second": 8.497968509367116,
-    "gpu_batches_per_second": 1.0263006004353032,
-    "producer_headroom": 8.28019442428731,
+    "producer_batch_seconds": 0.45435247570361526,
+    "producer_p95_seconds": 0.3222355859979871,
+    "gpu_update_seconds": 0.9634936211204441,
+    "producer_batches_per_second": 2.2009344143033203,
+    "gpu_batches_per_second": 1.037889590630712,
+    "producer_headroom": 2.12058626868571,
     "sustainable": true,
-    "buffer_seconds": 60.41115046966048,
+    "buffer_seconds": 60.70009813058798,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367079.0331841491,
-    "upload_bytes_per_second": 4682540.104899273,
+    "generation_bytes_per_second": 352618.9967261478,
+    "upload_bytes_per_second": 4609413.04626973,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.76661535876613,
+    "estimated_full_shards_per_hour": 75.66382814730001,
     "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": 1.5521136860006663,
-      "bytes": 11684886,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.2612271040015912,
       "bytes": 2529037,
@@ -562,6 +556,12 @@
       "bytes": 2354584,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.836964503992931,
+      "bytes": 6252743,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -390,3 +390,4 @@
 {"step": 799000, "loss": 0.49883708357810974, "lr": 3e-05, "grad_norm": 0.27126486284628265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116950482357, "tokens_per_second": 151958.34753286018, "data_wait_s": 1.377000444335863e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 799100, "loss": 0.4676346182823181, "lr": 3e-05, "grad_norm": 0.26486461381684856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116964977603, "tokens_per_second": 150731.913292813, "data_wait_s": 1.1967997124884278e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 799200, "loss": 0.5641132593154907, "lr": 3e-05, "grad_norm": 0.28709438999567377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116979520606, "tokens_per_second": 137407.55568114898, "data_wait_s": 1.556699862703681e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 799300, "loss": 0.5825047492980957, "lr": 3e-05, "grad_norm": 0.28232000321448114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116994211030, "tokens_per_second": 152123.7532338152, "data_wait_s": 2.0558996766339988e-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:e2e12178f10d4e575c09e50b56b96168d45a29d08bca3ba79e3c940eb735c0c4
-size 835902
+oid sha256:3eefcb427b1291fd8b05cb2fd022d4b35a7b038da19353f5f43fe3dc426a595e
+size 838020