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": 76927.15430241119,
-      "required_pages_per_second_upper_estimate": 0.005321999030230681
+      "required_token_positions_per_second": 76638.45451160955,
+      "required_pages_per_second_upper_estimate": 0.0053020261088272186
     },
     {
       "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": 9615.894287801399,
-      "required_pages_per_second_upper_estimate": 0.0040259741355956795
+      "required_token_positions_per_second": 9579.806813951194,
+      "required_pages_per_second_upper_estimate": 0.004010865063886746
     },
     {
       "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": 19231.788575602797,
-      "required_pages_per_second_upper_estimate": 0.004005474383927191
+      "required_token_positions_per_second": 19159.61362790239,
+      "required_pages_per_second_upper_estimate": 0.003990442245702563
     },
     {
       "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": 28847.682863404196,
-      "required_pages_per_second_upper_estimate": 0.001729374539799803
+      "required_token_positions_per_second": 28739.420441853585,
+      "required_pages_per_second_upper_estimate": 0.00172288437293499
     },
     {
       "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": 28847.682863404196,
-      "required_pages_per_second_upper_estimate": 0.0022427804550826493
+      "required_token_positions_per_second": 28739.420441853585,
+      "required_pages_per_second_upper_estimate": 0.002234363528003156
     },
     {
       "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": 28847.682863404196,
-      "required_pages_per_second_upper_estimate": 0.37002710154313306
+      "required_token_positions_per_second": 28739.420441853585,
+      "required_pages_per_second_upper_estimate": 0.36863842744261344
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.0688712199296333,
-    "producer_p95_seconds": 0.2343837699991127,
-    "gpu_update_seconds": 0.9584131984158042,
-    "producer_batches_per_second": 14.519853155232537,
-    "gpu_batches_per_second": 1.043391307270117,
-    "producer_headroom": 13.916018903034221,
+    "producer_batch_seconds": 0.1253286004685208,
+    "producer_p95_seconds": 0.2523967419983819,
+    "gpu_update_seconds": 0.9620235751078635,
+    "producer_batches_per_second": 7.979024709935809,
+    "gpu_batches_per_second": 1.0394755657499126,
+    "producer_headroom": 7.676009877326431,
     "sustainable": true,
-    "buffer_seconds": 60.38003150019566,
+    "buffer_seconds": 60.6074852317954,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 257745.6144786133,
-    "upload_bytes_per_second": 4897227.352556472,
+    "generation_bytes_per_second": 258699.48297762065,
+    "upload_bytes_per_second": 4864107.23145342,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.30620885628509,
+    "estimated_full_shards_per_hour": 55.510886831249856,
     "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": 1.1982196040007693,
-      "bytes": 8227634,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.881202160999237,
       "bytes": 6307877,
@@ -562,6 +556,12 @@
       "bytes": 5770251,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.3281948670010024,
+      "bytes": 5971602,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 461800, "loss": 0.5601227879524231, "lr": 0.0003, "grad_norm": 0.147232928934239, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67593688787, "tokens_per_second": 151329.49784767508, "data_wait_s": 1.136699938797392e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 461900, "loss": 0.5928270220756531, "lr": 0.0003, "grad_norm": 0.14946183798618587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67608199556, "tokens_per_second": 150451.15250701003, "data_wait_s": 1.2897999113192782e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 462000, "loss": 0.5891892313957214, "lr": 0.0003, "grad_norm": 0.14247113409333892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67622733900, "tokens_per_second": 151880.6014959981, "data_wait_s": 8.271999831777066e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 462100, "loss": 0.537992000579834, "lr": 0.0003, "grad_norm": 0.1823219599485683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67637430061, "tokens_per_second": 152737.53840355968, "data_wait_s": 8.814997272565961e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7f533c94d4159c4d85e76f0162bc7f33cb6a2f255d162f690e526515a5413675
-size 465021
+oid sha256:6c339196355605da5c2bc4e05b7b28491efcd615d86c9c62dcef7f93a2e4c503
+size 467139