Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75229.4140338592,
-      "required_pages_per_second_upper_estimate": 0.006520694941248211
+      "required_token_positions_per_second": 75232.74148682141,
+      "required_pages_per_second_upper_estimate": 0.006520983356437623
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9403.6767542324,
-      "required_pages_per_second_upper_estimate": 0.0037512672547600125
+      "required_token_positions_per_second": 9404.092685852676,
+      "required_pages_per_second_upper_estimate": 0.003751433176102073
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18807.3535084648,
-      "required_pages_per_second_upper_estimate": 0.004350750236066572
+      "required_token_positions_per_second": 18808.185371705353,
+      "required_pages_per_second_upper_estimate": 0.00435094267298698
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28211.0302626972,
-      "required_pages_per_second_upper_estimate": 0.0017179037311614287
+      "required_token_positions_per_second": 28212.27805755803,
+      "required_pages_per_second_upper_estimate": 0.0017179797153217794
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28211.0302626972,
-      "required_pages_per_second_upper_estimate": 0.00217176320275238
+      "required_token_positions_per_second": 28212.27805755803,
+      "required_pages_per_second_upper_estimate": 0.002171859261454884
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28211.0302626972,
-      "required_pages_per_second_upper_estimate": 0.350286579618029
+      "required_token_positions_per_second": 28212.27805755803,
+      "required_pages_per_second_upper_estimate": 0.3503020730540997
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.24345651046877492,
-    "producer_p95_seconds": 0.31245939200016437,
-    "gpu_update_seconds": 0.9800421942249418,
-    "producer_batches_per_second": 4.10750978922068,
-    "gpu_batches_per_second": 1.0203642311450087,
-    "producer_headroom": 4.025532906628263,
+    "producer_batch_seconds": 0.4707866918438981,
+    "producer_p95_seconds": 0.31213323500196566,
+    "gpu_update_seconds": 0.9799988481466544,
+    "producer_batches_per_second": 2.1241042224098736,
+    "gpu_batches_per_second": 1.0204093626142228,
+    "producer_headroom": 2.081619691305121,
     "sustainable": true,
-    "buffer_seconds": 60.76261604194639,
+    "buffer_seconds": 60.759928585092574,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493625.461825298,
-    "upload_bytes_per_second": 5194047.996868255,
+    "generation_bytes_per_second": 424751.4541867612,
+    "upload_bytes_per_second": 3943201.6653003013,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.92053309506612,
+    "estimated_full_shards_per_hour": 91.14177436067703,
     "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": 0.9793735439998272,
-      "bytes": 7348514,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1231975430018792,
       "bytes": 5800198,
@@ -562,6 +556,12 @@
       "bytes": 1082411,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 27.483964940998703,
+      "bytes": 11288612,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -256,3 +256,4 @@
 {"step": 340600, "loss": 0.5959947258234024, "lr": 0.0003, "grad_norm": 0.14344431628262702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49854778812, "tokens_per_second": 150351.13342953153, "data_wait_s": 1.4061002730159089e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340700, "loss": 0.6126907952129841, "lr": 0.0003, "grad_norm": 0.16114369116465935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49869487722, "tokens_per_second": 150051.98153942704, "data_wait_s": 1.6572001186432317e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340800, "loss": 0.6215404812246561, "lr": 0.0003, "grad_norm": 0.18114448725654078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49884120556, "tokens_per_second": 149051.0557850749, "data_wait_s": 1.3945002137916163e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 340900, "loss": 0.5959616415202618, "lr": 0.0003, "grad_norm": 0.14481558845554657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49898702209, "tokens_per_second": 148602.9770913688, "data_wait_s": 1.7546000890433788e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:38f7a5e4337d9e873e85d2bc87a2ab9d3e49b89c5d920c8a2d7212f1acf4b0a6
-size 550524
+oid sha256:5d1726bcc203ae2a5242bb7bfb9c537242cc950a807391eaf0125f294861f1a6
+size 552642