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": 76798.57509850581,
-      "required_pages_per_second_upper_estimate": 0.00531310362255965
+      "required_token_positions_per_second": 75477.72416598386,
+      "required_pages_per_second_upper_estimate": 0.005221724090251371
     },
     {
       "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": 9599.821887313226,
-      "required_pages_per_second_upper_estimate": 0.004019244957141169
+      "required_token_positions_per_second": 9434.715520747983,
+      "required_pages_per_second_upper_estimate": 0.0039501183692733
     },
     {
       "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": 19199.643774626453,
-      "required_pages_per_second_upper_estimate": 0.003998779469599226
+      "required_token_positions_per_second": 18869.431041495965,
+      "required_pages_per_second_upper_estimate": 0.003930004865583526
     },
     {
       "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": 28799.465661939677,
-      "required_pages_per_second_upper_estimate": 0.0017264839921954087
+      "required_token_positions_per_second": 28304.14656224395,
+      "required_pages_per_second_upper_estimate": 0.0016967903684771235
     },
     {
       "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": 28799.465661939677,
-      "required_pages_per_second_upper_estimate": 0.0022390317797538386
+      "required_token_positions_per_second": 28304.14656224395,
+      "required_pages_per_second_upper_estimate": 0.0022005228984309654
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28799.465661939677,
-      "required_pages_per_second_upper_estimate": 0.3694086230543435
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28304.14656224395,
+      "required_pages_per_second_upper_estimate": 0.3630552014756603
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07881572188283315,
-    "producer_p95_seconds": 0.21754720699937025,
-    "gpu_update_seconds": 0.9600178115991432,
-    "producer_batches_per_second": 12.68782390252788,
-    "gpu_batches_per_second": 1.041647340203258,
-    "producer_headroom": 12.180536936860117,
+    "producer_batch_seconds": 0.11659212689070841,
+    "producer_p95_seconds": 0.2283503800008475,
+    "gpu_update_seconds": 0.9768180057716628,
+    "producer_batches_per_second": 8.576908464302946,
+    "gpu_batches_per_second": 1.0237321528589391,
+    "producer_headroom": 8.378078621786498,
     "sustainable": true,
-    "buffer_seconds": 60.48112213074602,
+    "buffer_seconds": 60.562716357843094,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445007.79494910984,
-    "upload_bytes_per_second": 5219976.993085824,
+    "generation_bytes_per_second": 460329.1437808124,
+    "upload_bytes_per_second": 5202698.920922406,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.48831354479762,
+    "estimated_full_shards_per_hour": 98.77591834133413,
     "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": 0.993601779000528,
-      "bytes": 1093689,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0732885580000584,
       "bytes": 7199351,
@@ -562,6 +556,12 @@
       "bytes": 11659745,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.587534140999196,
+      "bytes": 2763901,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -161,3 +161,4 @@
 {"step": 776100, "loss": 0.5250530242919922, "lr": 3e-05, "grad_norm": 0.2389914929713818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113598928558, "tokens_per_second": 152751.50531584598, "data_wait_s": 1.7178001144202426e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 776200, "loss": 0.5679916143417358, "lr": 3e-05, "grad_norm": 0.2680121779115453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113613538215, "tokens_per_second": 151672.56654686952, "data_wait_s": 1.1180000001331791e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 776300, "loss": 0.5306581258773804, "lr": 3e-05, "grad_norm": 0.2461960905525235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113628204666, "tokens_per_second": 151262.90834001784, "data_wait_s": 1.1126001481898129e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 776400, "loss": 0.5531155467033386, "lr": 3e-05, "grad_norm": 0.2534336618671871, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113642821074, "tokens_per_second": 151769.46908746462, "data_wait_s": 1.591500040376559e-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:b7a37cd16d3bdca0d247e24ff6ffc69d52a4a6ce9aa10c3a059ce5982b98410b
-size 347748
+oid sha256:27ed9d69e4e03e55d2c7ae1c4726681a949a4b2e65e469615abe5075eb400159
+size 349866