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": 76418.5528273331,
-      "required_pages_per_second_upper_estimate": 0.005286812800066768
+      "required_token_positions_per_second": 76646.04657194113,
+      "required_pages_per_second_upper_estimate": 0.005302551345176963
     },
     {
       "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": 9552.319103416638,
-      "required_pages_per_second_upper_estimate": 0.0039993565334945965
+      "required_token_positions_per_second": 9580.755821492641,
+      "required_pages_per_second_upper_estimate": 0.0040112623935268195
     },
     {
       "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": 19104.638206833275,
-      "required_pages_per_second_upper_estimate": 0.003978992315293215
+      "required_token_positions_per_second": 19161.511642985282,
+      "required_pages_per_second_upper_estimate": 0.003990837552190306
     },
     {
       "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": 28656.957310249913,
-      "required_pages_per_second_upper_estimate": 0.001717940834109261
+      "required_token_positions_per_second": 28742.26746447792,
+      "required_pages_per_second_upper_estimate": 0.0017230550475941671
     },
     {
       "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": 28656.957310249913,
-      "required_pages_per_second_upper_estimate": 0.002227952382237953
+      "required_token_positions_per_second": 28742.26746447792,
+      "required_pages_per_second_upper_estimate": 0.002234584871489469
     },
     {
       "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": 28656.957310249913,
-      "required_pages_per_second_upper_estimate": 0.3675806789324138
+      "required_token_positions_per_second": 28742.26746447792,
+      "required_pages_per_second_upper_estimate": 0.36867494599194367
     }
   ],
   "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": 0.12129860746838972,
-    "producer_p95_seconds": 0.2620791130029829,
-    "gpu_update_seconds": 0.9647918898253364,
-    "producer_batches_per_second": 8.24411772625336,
-    "gpu_batches_per_second": 1.0364929582700344,
-    "producer_headroom": 7.953857921054535,
+    "producer_batch_seconds": 0.07118791676552405,
+    "producer_p95_seconds": 0.2464940559984825,
+    "gpu_update_seconds": 0.9619282832911388,
+    "producer_batches_per_second": 14.04732776903362,
+    "gpu_batches_per_second": 1.0395785396584896,
+    "producer_headroom": 13.512521885694452,
     "sustainable": true,
-    "buffer_seconds": 60.78188905899619,
+    "buffer_seconds": 60.601481847341745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 302502.0164648046,
-    "upload_bytes_per_second": 4980481.402898315,
+    "generation_bytes_per_second": 320343.3805060409,
+    "upload_bytes_per_second": 4994627.412620945,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.90989084680656,
+    "estimated_full_shards_per_hour": 68.73823224435729,
     "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": 1.2100102580006933,
-      "bytes": 5596380,
-      "pages": 21,
-      "shards": 2
-    },
     {
       "seconds": 1.2646026780003012,
       "bytes": 13750693,
@@ -562,6 +556,12 @@
       "bytes": 2627126,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0633560690002923,
+      "bytes": 6091672,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -201,3 +201,4 @@
 {"step": 780100, "loss": 0.5658770799636841, "lr": 3e-05, "grad_norm": 0.2438217714019965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114184617151, "tokens_per_second": 145341.33308708438, "data_wait_s": 9.096002031583339e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 780200, "loss": 0.5623610615730286, "lr": 3e-05, "grad_norm": 0.24890262904165042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114199317450, "tokens_per_second": 152029.68073092971, "data_wait_s": 9.111998224398121e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 780300, "loss": 0.506441593170166, "lr": 3e-05, "grad_norm": 0.25383359452171794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114213963840, "tokens_per_second": 152177.9096694519, "data_wait_s": 1.300100120715797e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 780400, "loss": 0.5481046438217163, "lr": 3e-05, "grad_norm": 0.26408582832388844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114228543603, "tokens_per_second": 152106.15511811947, "data_wait_s": 9.294999472331256e-06, "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 sha256:15f55238916ad98d05df2ab4591e2df0332e478a0c911a022079d695b6e12974
-size 433251
+oid sha256:1f42344aff837bc8fe432319f4f7fc9515263f6cd7de79ff1dc5f23139392a27
+size 435369