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": 75930.66274117952,
-      "required_pages_per_second_upper_estimate": 0.0052530594318456955
+      "required_token_positions_per_second": 76343.45874872459,
+      "required_pages_per_second_upper_estimate": 0.0052816176174663625
     },
     {
       "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": 9491.33284264744,
-      "required_pages_per_second_upper_estimate": 0.003973822859648477
+      "required_token_positions_per_second": 9542.932343590574,
+      "required_pages_per_second_upper_estimate": 0.003995426493173258
     },
     {
       "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": 18982.66568529488,
-      "required_pages_per_second_upper_estimate": 0.003953588655688469
+      "required_token_positions_per_second": 19085.864687181147,
+      "required_pages_per_second_upper_estimate": 0.0039750822862406835
     },
     {
       "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": 28473.99852794232,
-      "required_pages_per_second_upper_estimate": 0.001706972734471806
+      "required_token_positions_per_second": 28628.79703077172,
+      "required_pages_per_second_upper_estimate": 0.0017162526683528005
     },
     {
       "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": 28473.99852794232,
-      "required_pages_per_second_upper_estimate": 0.0022137281416641725
+      "required_token_positions_per_second": 28628.79703077172,
+      "required_pages_per_second_upper_estimate": 0.0022257630443725026
     },
     {
       "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": 28473.99852794232,
-      "required_pages_per_second_upper_estimate": 0.3652338801188071
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28628.79703077172,
+      "required_pages_per_second_upper_estimate": 0.3672194691034199
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.10179067010157539,
-    "producer_p95_seconds": 0.15287248399999953,
-    "gpu_update_seconds": 0.9709911297799729,
-    "producer_batches_per_second": 9.824083081505554,
-    "gpu_batches_per_second": 1.0298755254608767,
-    "producer_headroom": 9.539097530363396,
+    "producer_batch_seconds": 0.04642860108594604,
+    "producer_p95_seconds": 0.1571371850000105,
+    "gpu_update_seconds": 0.9657408926502393,
+    "producer_batches_per_second": 21.538447780256305,
+    "gpu_batches_per_second": 1.0354744296430742,
+    "producer_headroom": 20.80055978560529,
     "sustainable": true,
-    "buffer_seconds": 60.201450046358325,
+    "buffer_seconds": 60.84167623696508,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 449374.8202830633,
-    "upload_bytes_per_second": 4116576.3146626665,
+    "generation_bytes_per_second": 449220.1601039408,
+    "upload_bytes_per_second": 4158056.942373036,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.42537552231715,
+    "estimated_full_shards_per_hour": 96.39218904818219,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 11417123,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3576498049997099,
+      "bytes": 7486146,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -34,3 +34,4 @@
 {"step": 845900, "loss": 0.5151355862617493, "lr": 3e-05, "grad_norm": 0.2853149347511223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123816069731, "tokens_per_second": 153360.16995992142, "data_wait_s": 9.447000138607109e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 846000, "loss": 0.5326921343803406, "lr": 3e-05, "grad_norm": 0.2937797453512959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123830789717, "tokens_per_second": 152965.5423731737, "data_wait_s": 1.2217999938002322e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 846100, "loss": 0.5296584367752075, "lr": 3e-05, "grad_norm": 0.298873914583901, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123845504747, "tokens_per_second": 153350.81891504614, "data_wait_s": 9.380999927088851e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 846200, "loss": 0.5164526104927063, "lr": 3e-05, "grad_norm": 0.31853155076772827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123860162102, "tokens_per_second": 152481.32985085127, "data_wait_s": 1.1221000022487715e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1ce6cb3bf3896fc5e6f83b5cac6ee9ad13ec77fd132a58af16b7e7723d4720ea
-size 77196
+oid sha256:17f2f1e3e76f30a44e5669621a90218656b16d25145cde98642d67534d5d3d46
+size 79314