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": 75747.96771263136,
-      "required_pages_per_second_upper_estimate": 0.005240420165859864
+      "required_token_positions_per_second": 74917.44209604662,
+      "required_pages_per_second_upper_estimate": 0.005182962476619602
     },
     {
       "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": 9468.49596407892,
-      "required_pages_per_second_upper_estimate": 0.003964261535480091
+      "required_token_positions_per_second": 9364.680262005828,
+      "required_pages_per_second_upper_estimate": 0.003920796068940469
     },
     {
       "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": 18936.99192815784,
-      "required_pages_per_second_upper_estimate": 0.003944076016574798
+      "required_token_positions_per_second": 18729.360524011656,
+      "required_pages_per_second_upper_estimate": 0.0039008318706994943
     },
     {
       "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": 28405.48789223676,
-      "required_pages_per_second_upper_estimate": 0.0017028656264709413
+      "required_token_positions_per_second": 28094.040786017486,
+      "required_pages_per_second_upper_estimate": 0.0016841948480052916
     },
     {
       "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": 28405.48789223676,
-      "required_pages_per_second_upper_estimate": 0.0022084017410845046
+      "required_token_positions_per_second": 28094.040786017486,
+      "required_pages_per_second_upper_estimate": 0.0021841880984869995
     },
     {
       "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": 28405.48789223676,
-      "required_pages_per_second_upper_estimate": 0.36435509924496556
+      "required_token_positions_per_second": 28094.040786017486,
+      "required_pages_per_second_upper_estimate": 0.36036019017223336
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11652121721093067,
-    "producer_p95_seconds": 0.32131955699878745,
-    "gpu_update_seconds": 0.9733330441247664,
-    "producer_batches_per_second": 8.582127992962569,
-    "gpu_batches_per_second": 1.0273975655467578,
-    "producer_headroom": 8.35326876445863,
+    "producer_batch_seconds": 0.49265519849996053,
+    "producer_p95_seconds": 0.3395640140006435,
+    "gpu_update_seconds": 0.9841232954200209,
+    "producer_batches_per_second": 2.0298172089623856,
+    "gpu_batches_per_second": 1.0161328409294519,
+    "producer_headroom": 1.9975904007843321,
     "sustainable": true,
-    "buffer_seconds": 60.34664873573551,
+    "buffer_seconds": 60.03152102062127,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373502.47145661653,
-    "upload_bytes_per_second": 5579985.104406778,
+    "generation_bytes_per_second": 330491.498289128,
+    "upload_bytes_per_second": 5614715.235665415,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.14493568204757,
+    "estimated_full_shards_per_hour": 70.91578208451634,
     "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": 1.0211126419999346,
-      "bytes": 2557759,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.475009352001507,
       "bytes": 6011684,
@@ -562,6 +556,12 @@
       "bytes": 6839680,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1501934409971,
+      "bytes": 5967653,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -330,3 +330,4 @@
 {"step": 430100, "loss": 0.566529393196106, "lr": 0.0003, "grad_norm": 0.15416684422087926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62954780166, "tokens_per_second": 138170.1454360266, "data_wait_s": 1.3216002116678283e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 430200, "loss": 0.5586528182029724, "lr": 0.0003, "grad_norm": 0.1566217699200918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62969432541, "tokens_per_second": 151790.91207930402, "data_wait_s": 1.2141004845034331e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 430300, "loss": 0.5544184446334839, "lr": 0.0003, "grad_norm": 0.17044130918967768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62984072687, "tokens_per_second": 151063.73952058694, "data_wait_s": 1.5867997717577964e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 430400, "loss": 0.5803173184394836, "lr": 0.0003, "grad_norm": 0.15600094089472244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62998840169, "tokens_per_second": 153888.17733639732, "data_wait_s": 7.443995855282992e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b9c7f84f7a5439e5ad997512d1bc2ef177ba12239fac2e2837a485399f80e86
-size 708822
+oid sha256:af028096540557fab7bfc7c99f2e12f72d2c98da77b90ae68fe33b9114bc7c87
+size 710940