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": 75475.5883545632,
-      "required_pages_per_second_upper_estimate": 0.006542032707443721
+      "required_token_positions_per_second": 75069.90335934206,
+      "required_pages_per_second_upper_estimate": 0.006506868960257174
     },
     {
       "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": 9434.4485443204,
-      "required_pages_per_second_upper_estimate": 0.003763542581905377
+      "required_token_positions_per_second": 9383.737919917758,
+      "required_pages_per_second_upper_estimate": 0.0037433133556397627
     },
     {
       "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": 18868.8970886408,
-      "required_pages_per_second_upper_estimate": 0.004364987260210272
+      "required_token_positions_per_second": 18767.475839835515,
+      "required_pages_per_second_upper_estimate": 0.004341525239252185
     },
     {
       "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": 28303.345632961198,
-      "required_pages_per_second_upper_estimate": 0.00172352525286919
+      "required_token_positions_per_second": 28151.21375975327,
+      "required_pages_per_second_upper_estimate": 0.00171426121996508
     },
     {
       "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": 28303.345632961198,
-      "required_pages_per_second_upper_estimate": 0.0021788698955006033
+      "required_token_positions_per_second": 28151.21375975327,
+      "required_pages_per_second_upper_estimate": 0.0021671583627730787
     },
     {
       "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": 28303.345632961198,
-      "required_pages_per_second_upper_estimate": 0.35143282755703836
+      "required_token_positions_per_second": 28151.21375975327,
+      "required_pages_per_second_upper_estimate": 0.3495438588444227
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10494868408611069,
-    "producer_p95_seconds": 0.34983729099985794,
-    "gpu_update_seconds": 0.9768456478092822,
-    "producer_batches_per_second": 9.528466304346392,
-    "gpu_batches_per_second": 1.0237031840625435,
-    "producer_headroom": 9.307840839698168,
+    "producer_batch_seconds": 0.5781725355701894,
+    "producer_p95_seconds": 0.37051739700109465,
+    "gpu_update_seconds": 0.9821246158674445,
+    "producer_batches_per_second": 1.7295875166637023,
+    "gpu_batches_per_second": 1.0182007291577428,
+    "producer_headroom": 1.6986704754124657,
     "sustainable": true,
-    "buffer_seconds": 60.5644301641755,
+    "buffer_seconds": 60.89172618378156,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291062.7391103045,
-    "upload_bytes_per_second": 2810096.625182081,
+    "generation_bytes_per_second": 286925.25603695895,
+    "upload_bytes_per_second": 2833233.3178105312,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.45528821927883,
+    "estimated_full_shards_per_hour": 61.567480667415396,
     "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": 1.2782237920000625,
-      "bytes": 7965202,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.4521252480008116,
       "bytes": 10591031,
@@ -562,6 +556,12 @@
       "bytes": 6125090,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1863440980014275,
+      "bytes": 11209339,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -281,3 +281,4 @@
 {"step": 223100, "loss": 0.6118058785796165, "lr": 0.0003, "grad_norm": 0.13831426155139062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32654974765, "tokens_per_second": 148573.64504332296, "data_wait_s": 1.252899892278947e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223200, "loss": 0.6013508550822735, "lr": 0.0003, "grad_norm": 0.16874810110957425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32669695844, "tokens_per_second": 149911.66282811202, "data_wait_s": 1.692300065769814e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 223300, "loss": 0.5461439806967974, "lr": 0.0003, "grad_norm": 0.1377671121957184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32684475373, "tokens_per_second": 137166.45368893736, "data_wait_s": 1.1956002708757296e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 223400, "loss": 0.552930623292923, "lr": 0.0003, "grad_norm": 0.156365045295204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32698972515, "tokens_per_second": 147221.9307159658, "data_wait_s": 1.4793000445934013e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b5ca43d7ff29c8eeb10336f229f5be5ec15128890fe2b9793ca4754a8ba6b53
-size 603474
+oid sha256:6b58213fa4fb0c65ab429871b5742fe0eb59ef4cc830f383c85063bb93b46cf1
+size 605592