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": 76809.17316697267,
-      "required_pages_per_second_upper_estimate": 0.005313836821527094
+      "required_token_positions_per_second": 76501.72505532067,
+      "required_pages_per_second_upper_estimate": 0.005292566847785105
     },
     {
       "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": 9601.146645871584,
-      "required_pages_per_second_upper_estimate": 0.0040197996058854496
+      "required_token_positions_per_second": 9562.715631915084,
+      "required_pages_per_second_upper_estimate": 0.004003709342872693
     },
     {
       "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": 19202.293291743168,
-      "required_pages_per_second_upper_estimate": 0.0039993312941421725
+      "required_token_positions_per_second": 19125.431263830167,
+      "required_pages_per_second_upper_estimate": 0.003983322960715879
     },
     {
       "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": 28803.439937614752,
-      "required_pages_per_second_upper_estimate": 0.001726722243953762
+      "required_token_positions_per_second": 28688.146895745253,
+      "required_pages_per_second_upper_estimate": 0.0017198106021359672
     },
     {
       "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": 28803.439937614752,
-      "required_pages_per_second_upper_estimate": 0.0022393407622065846
+      "required_token_positions_per_second": 28688.146895745253,
+      "required_pages_per_second_upper_estimate": 0.0022303772353217278
     },
     {
       "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": 28803.439937614752,
-      "required_pages_per_second_upper_estimate": 0.36945960079545864
+      "required_token_positions_per_second": 28688.146895745253,
+      "required_pages_per_second_upper_estimate": 0.36798074544638026
     }
   ],
   "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": 0.11367300503940214,
-    "producer_p95_seconds": 0.3086689909978304,
-    "gpu_update_seconds": 0.9598853491070055,
-    "producer_batches_per_second": 8.797163404392915,
-    "gpu_batches_per_second": 1.041791085706552,
-    "producer_headroom": 8.444268265577067,
+    "producer_batch_seconds": 0.11909220940651721,
+    "producer_p95_seconds": 0.320237982996332,
+    "gpu_update_seconds": 0.9637429737262145,
+    "producer_batches_per_second": 8.396854882308329,
+    "gpu_batches_per_second": 1.0376210538102304,
+    "producer_headroom": 8.092409894223312,
     "sustainable": true,
-    "buffer_seconds": 60.47277699374135,
+    "buffer_seconds": 60.71580734475152,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380519.6126083594,
-    "upload_bytes_per_second": 4913891.36291536,
+    "generation_bytes_per_second": 369301.39925219084,
+    "upload_bytes_per_second": 4891599.428851855,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.65065082252586,
+    "estimated_full_shards_per_hour": 79.24348338293356,
     "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": 1.0191816589976952,
-      "bytes": 7135611,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1397755830003007,
       "bytes": 6913587,
@@ -562,6 +556,12 @@
       "bytes": 6059365,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.2079227729991544,
+      "bytes": 6092047,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -360,3 +360,4 @@
 {"step": 640400, "loss": 0.5551453232765198, "lr": 0.0001, "grad_norm": 0.16150456430865168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93738132504, "tokens_per_second": 151911.8700938095, "data_wait_s": 1.0700001439545304e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 640500, "loss": 0.5217152237892151, "lr": 0.0001, "grad_norm": 0.1658666262688848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93752800717, "tokens_per_second": 152543.14439320512, "data_wait_s": 9.162002243101597e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 640600, "loss": 0.5625096559524536, "lr": 0.0001, "grad_norm": 0.19436205433084347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93767531675, "tokens_per_second": 153343.26154659578, "data_wait_s": 1.4709003153257072e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 640700, "loss": 0.5348134636878967, "lr": 0.0001, "grad_norm": 0.1798952781332512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93782216773, "tokens_per_second": 153296.59126634218, "data_wait_s": 1.3729004422202706e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:53da49e1bae342117ab65d8693a2ac92b654e29b3767004e6b23ae4c24c6e3e1
-size 773145
+oid sha256:b09ceade3661b791dfe8bf83d41a091f3a1cfb8ff3384f79a2171b359bea83cf
+size 775263