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": 75210.7461503121,
-      "required_pages_per_second_upper_estimate": 0.006519076856415667
+      "required_token_positions_per_second": 75171.35175096683,
+      "required_pages_per_second_upper_estimate": 0.006515662249724611
     },
     {
       "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": 9401.343268789013,
-      "required_pages_per_second_upper_estimate": 0.003750336392527929
+      "required_token_positions_per_second": 9396.418968870854,
+      "required_pages_per_second_upper_estimate": 0.0037483720156657307
     },
     {
       "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": 18802.686537578025,
-      "required_pages_per_second_upper_estimate": 0.004349670614487799
+      "required_token_positions_per_second": 18792.837937741708,
+      "required_pages_per_second_upper_estimate": 0.004347392314245104
     },
     {
       "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": 28204.029806367038,
-      "required_pages_per_second_upper_estimate": 0.0017174774400994775
+      "required_token_positions_per_second": 28189.25690661256,
+      "required_pages_per_second_upper_estimate": 0.001716577848011845
     },
     {
       "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": 28204.029806367038,
-      "required_pages_per_second_upper_estimate": 0.002171224288245581
+      "required_token_positions_per_second": 28189.25690661256,
+      "required_pages_per_second_upper_estimate": 0.0021700870295284776
     },
     {
       "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": 28204.029806367038,
-      "required_pages_per_second_upper_estimate": 0.35019965737942854
+      "required_token_positions_per_second": 28189.25690661256,
+      "required_pages_per_second_upper_estimate": 0.3500162274061929
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10185332225790944,
-    "producer_p95_seconds": 0.3409251650009537,
-    "gpu_update_seconds": 0.9802854482077765,
-    "producer_batches_per_second": 9.818040077944977,
-    "gpu_batches_per_second": 1.0201110317696411,
-    "producer_headroom": 9.624481818330205,
+    "producer_batch_seconds": 0.13291685907830697,
+    "producer_p95_seconds": 0.36266010999679565,
+    "gpu_update_seconds": 0.9807991779135691,
+    "producer_batches_per_second": 7.5235000806847045,
+    "gpu_batches_per_second": 1.0195767110319938,
+    "producer_headroom": 7.379042694168229,
     "sustainable": true,
-    "buffer_seconds": 60.77769778888214,
+    "buffer_seconds": 60.809549030641286,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 275812.90667042736,
-    "upload_bytes_per_second": 5724593.863587075,
+    "generation_bytes_per_second": 280627.65930957696,
+    "upload_bytes_per_second": 5688091.940583648,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.183029175611644,
+    "estimated_full_shards_per_hour": 60.21616301026803,
     "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": 1.6663710430002538,
-      "bytes": 11892665,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.4550072420024662,
       "bytes": 2481838,
@@ -562,6 +556,12 @@
       "bytes": 10955791,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.4038716970026144,
+      "bytes": 7663039,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 306400, "loss": 0.5626558177173138, "lr": 0.0003, "grad_norm": 0.14518262905942703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44848651720, "tokens_per_second": 149997.13231858963, "data_wait_s": 1.140800304710865e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 306500, "loss": 0.5580325722694397, "lr": 0.0003, "grad_norm": 0.15709238989062616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44863294944, "tokens_per_second": 150351.97776786666, "data_wait_s": 8.226001227740198e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 306600, "loss": 0.6092037744820118, "lr": 0.0003, "grad_norm": 0.1592246802965326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44877971060, "tokens_per_second": 149919.76741815222, "data_wait_s": 1.1723001080099493e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 306700, "loss": 0.5748039167374372, "lr": 0.0003, "grad_norm": 0.14108368454578674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44892599415, "tokens_per_second": 150016.32332066845, "data_wait_s": 1.2319003872107714e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c648239834d82c4d7884bd5efbfc6932c4ed497f1155e8ae2fcfad6dfd8e4f67
-size 563232
+oid sha256:cf0eca2725b63af4279681599bbc0f12d278cd0a6c9a85b1c09a69430b5f395b
+size 565350