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": 77070.43158003519,
-      "required_pages_per_second_upper_estimate": 0.0053319112847457955
+      "required_token_positions_per_second": 76740.21498940256,
+      "required_pages_per_second_upper_estimate": 0.005309066134797777
     },
     {
       "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": 9633.803947504399,
-      "required_pages_per_second_upper_estimate": 0.004033472536117103
+      "required_token_positions_per_second": 9592.52687367532,
+      "required_pages_per_second_upper_estimate": 0.004016190687268186
     },
     {
       "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": 19267.607895008798,
-      "required_pages_per_second_upper_estimate": 0.004012934603540484
+      "required_token_positions_per_second": 19185.05374735064,
+      "required_pages_per_second_upper_estimate": 0.003995740751682568
     },
     {
       "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": 28901.4118425132,
-      "required_pages_per_second_upper_estimate": 0.001732595510057987
+      "required_token_positions_per_second": 28777.58062102596,
+      "required_pages_per_second_upper_estimate": 0.0017251720174091543
     },
     {
       "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": 28901.4118425132,
-      "required_pages_per_second_upper_estimate": 0.0022469576468795716
+      "required_token_positions_per_second": 28777.58062102596,
+      "required_pages_per_second_upper_estimate": 0.0022373303140849187
     },
     {
       "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": 28901.4118425132,
-      "required_pages_per_second_upper_estimate": 0.37071627919746025
+      "required_token_positions_per_second": 28777.58062102596,
+      "required_pages_per_second_upper_estimate": 0.36912790524782857
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09094359758591963,
+    "producer_batch_seconds": 0.11264592677329688,
     "producer_p95_seconds": 0.24977909400331555,
-    "gpu_update_seconds": 0.9566314666790962,
-    "producer_batches_per_second": 10.995826276338393,
-    "gpu_batches_per_second": 1.0453346297205295,
-    "producer_headroom": 10.518953418082141,
+    "gpu_update_seconds": 0.9607478948316924,
+    "producer_batches_per_second": 8.87737380875323,
+    "gpu_batches_per_second": 1.0408557805637284,
+    "producer_headroom": 8.528918198393669,
     "sustainable": true,
-    "buffer_seconds": 60.267782400783055,
+    "buffer_seconds": 60.527117374396624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337241.253587939,
-    "upload_bytes_per_second": 4932306.221605968,
+    "generation_bytes_per_second": 336177.9899655554,
+    "upload_bytes_per_second": 4785267.203027469,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.36412244537952,
+    "estimated_full_shards_per_hour": 72.13597082352635,
     "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": 1.8041386129989405,
-      "bytes": 11898323,
-      "pages": 40,
-      "shards": 4
-    },
-    {
-      "seconds": 1.2189035779992992,
-      "bytes": 6954573,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.5599068500014255,
       "bytes": 2415324,
@@ -562,6 +550,18 @@
       "bytes": 10380594,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1230385859998933,
+      "bytes": 790073,
+      "pages": 26,
+      "shards": 1
+    },
+    {
+      "seconds": 1.2272778379992815,
+      "bytes": 1850276,
+      "pages": 12,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -248,3 +248,4 @@
 {"step": 784800, "loss": 0.5423645377159119, "lr": 3e-05, "grad_norm": 0.2533305691222203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114872383127, "tokens_per_second": 149908.45026669605, "data_wait_s": 9.030001820065081e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 784900, "loss": 0.5186707973480225, "lr": 3e-05, "grad_norm": 0.25982937823500113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114886902437, "tokens_per_second": 150865.83335429293, "data_wait_s": 1.355299900751561e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 785000, "loss": 0.5375571846961975, "lr": 3e-05, "grad_norm": 0.25487106052843767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114901539099, "tokens_per_second": 151305.61949448384, "data_wait_s": 1.808800152502954e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 785100, "loss": 0.5449207425117493, "lr": 3e-05, "grad_norm": 0.2577403766505476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114916242606, "tokens_per_second": 145376.3739659378, "data_wait_s": 1.284600148210302e-05, "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 sha256:961e03f48bf8962f9768e1c12312d26f5208fd9243be6e8f2affa806d7d269ec
-size 533580
+oid sha256:54bee75cfd3ba73a475115dffb84e4a6df5e54b1554c6ddb69ed33f606848966
+size 535698