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": 76800.12522350495,
-      "required_pages_per_second_upper_estimate": 0.005313210863803875
+      "required_token_positions_per_second": 76776.36378229492,
+      "required_pages_per_second_upper_estimate": 0.005311566992166826
     },
     {
       "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": 9600.015652938118,
-      "required_pages_per_second_upper_estimate": 0.004019326082762026
+      "required_token_positions_per_second": 9597.045472786866,
+      "required_pages_per_second_upper_estimate": 0.004018082530357111
     },
     {
       "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": 19200.031305876237,
-      "required_pages_per_second_upper_estimate": 0.0039988601821386695
+      "required_token_positions_per_second": 19194.09094557373,
+      "required_pages_per_second_upper_estimate": 0.003997622961745493
     },
     {
       "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": 28800.046958814353,
-      "required_pages_per_second_upper_estimate": 0.0017265188400554566
+      "required_token_positions_per_second": 28791.1364183606,
+      "required_pages_per_second_upper_estimate": 0.0017259846667608617
     },
     {
       "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": 28800.046958814353,
-      "required_pages_per_second_upper_estimate": 0.00223907697302899
+      "required_token_positions_per_second": 28791.1364183606,
+      "required_pages_per_second_upper_estimate": 0.0022383842177020356
     },
     {
       "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": 28800.046958814353,
-      "required_pages_per_second_upper_estimate": 0.36941607930650394
+      "required_token_positions_per_second": 28791.1364183606,
+      "required_pages_per_second_upper_estimate": 0.3693017844609561
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.15711887622657628,
-    "producer_p95_seconds": 0.29682700900229975,
-    "gpu_update_seconds": 0.9599984347087405,
-    "producer_batches_per_second": 6.36460763987346,
-    "gpu_batches_per_second": 1.0416683651191534,
-    "producer_headroom": 6.110013371813813,
+    "producer_batch_seconds": 0.09796858180470736,
+    "producer_p95_seconds": 0.2911581770022167,
+    "gpu_update_seconds": 0.9602955436787969,
+    "producer_batches_per_second": 10.207354047376343,
+    "gpu_batches_per_second": 1.0413460799464915,
+    "producer_headroom": 9.802076604447233,
     "sustainable": true,
-    "buffer_seconds": 60.47990138665065,
+    "buffer_seconds": 60.498619251764204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315964.34629614296,
-    "upload_bytes_per_second": 4674474.790950291,
+    "generation_bytes_per_second": 297102.9072129166,
+    "upload_bytes_per_second": 4664345.447290908,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.7985934416124,
+    "estimated_full_shards_per_hour": 63.75136768618224,
     "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": 1.2818151600004057,
-      "bytes": 7876217,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.1551925370004028,
       "bytes": 6771836,
@@ -562,6 +556,12 @@
       "bytes": 10655327,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.3922439659982047,
+      "bytes": 7460601,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -291,3 +291,4 @@
 {"step": 871600, "loss": 0.508460521697998, "lr": 3e-05, "grad_norm": 0.2952682373814299, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127577619184, "tokens_per_second": 117957.411074345, "data_wait_s": 8.72599775902927e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 871700, "loss": 0.5211670398712158, "lr": 3e-05, "grad_norm": 0.300449272457701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127592222603, "tokens_per_second": 151054.1694849696, "data_wait_s": 1.438899926142767e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 871800, "loss": 0.6195017695426941, "lr": 3e-05, "grad_norm": 0.36921799767350544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127606975044, "tokens_per_second": 153495.28849010146, "data_wait_s": 1.8030001228908077e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 871900, "loss": 0.5731968879699707, "lr": 3e-05, "grad_norm": 0.34762593455913327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127621718746, "tokens_per_second": 152841.65976966496, "data_wait_s": 1.4327000826597214e-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 sha256:1098c6d909b7a9e9734eef9013ff0c43b0140b5288be14d1170aa2466d575c79
-size 625437
+oid sha256:91f53bfb5c8f613430658bdcbe90c0f5509d11b031e4095bbd7a02cf66636135
+size 627555