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": 75970.95974398145,
-      "required_pages_per_second_upper_estimate": 0.006584970243971188
+      "required_token_positions_per_second": 76139.20154465883,
+      "required_pages_per_second_upper_estimate": 0.006599553016848958
     },
     {
       "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": 9496.369967997682,
-      "required_pages_per_second_upper_estimate": 0.0037882439636180315
+      "required_token_positions_per_second": 9517.400193082354,
+      "required_pages_per_second_upper_estimate": 0.0037966332348341927
     },
     {
       "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": 18992.739935995363,
-      "required_pages_per_second_upper_estimate": 0.004393636123386078
+      "required_token_positions_per_second": 19034.80038616471,
+      "required_pages_per_second_upper_estimate": 0.004403366068294115
     },
     {
       "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": 28489.10990399305,
-      "required_pages_per_second_upper_estimate": 0.0017348373223452231
+      "required_token_positions_per_second": 28552.200579247063,
+      "required_pages_per_second_upper_estimate": 0.0017386792134570013
     },
     {
       "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": 28489.10990399305,
-      "required_pages_per_second_upper_estimate": 0.0021931705433130494
+      "required_token_positions_per_second": 28552.200579247063,
+      "required_pages_per_second_upper_estimate": 0.0021980274381402706
     },
     {
       "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": 28489.10990399305,
-      "required_pages_per_second_upper_estimate": 0.3537393980902324
+      "required_token_positions_per_second": 28552.200579247063,
+      "required_pages_per_second_upper_estimate": 0.35452277312598013
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05451448550782345,
-    "producer_p95_seconds": 0.1898002190000625,
-    "gpu_update_seconds": 0.970476090449033,
-    "producer_batches_per_second": 18.343748284233346,
-    "gpu_batches_per_second": 1.0304220885414153,
-    "producer_headroom": 17.802169119063933,
+    "producer_batch_seconds": 0.06856493126563379,
+    "producer_p95_seconds": 0.18790973399973154,
+    "gpu_update_seconds": 0.9683316675806672,
+    "producer_batches_per_second": 14.584715269760963,
+    "gpu_batches_per_second": 1.0327040140063317,
+    "producer_headroom": 14.122841658356855,
     "sustainable": true,
-    "buffer_seconds": 60.16951760784005,
+    "buffer_seconds": 60.03656339000137,
     "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": 401790.91924334905,
-    "upload_bytes_per_second": 4849503.360208995,
+    "generation_bytes_per_second": 415522.79566770466,
+    "upload_bytes_per_second": 4855432.1408156715,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.21497805571893,
+    "estimated_full_shards_per_hour": 89.16151907466273,
     "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": [
@@ -286,6 +286,12 @@
       "bytes": 2088001,
       "pages": 18,
       "shards": 2
+    },
+    {
+      "seconds": 1.2336509919996388,
+      "bytes": 6127983,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 160000, "loss": 0.5920981746166945, "lr": 0.0003, "grad_norm": 0.1573822250209816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23417085332, "tokens_per_second": 148750.36305334992, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 160100, "loss": 0.5946155153214931, "lr": 0.0003, "grad_norm": 0.14926821951092545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23431616199, "tokens_per_second": 136174.93388581954, "data_wait_s": 1.170799987448845e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 160200, "loss": 0.6043591499328613, "lr": 0.0003, "grad_norm": 0.14702501515498978, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23446262115, "tokens_per_second": 149684.11786736164, "data_wait_s": 1.3070999557385221e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 160300, "loss": 0.6843377538025379, "lr": 0.0003, "grad_norm": 0.1873446259741128, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23460885944, "tokens_per_second": 149292.79984657269, "data_wait_s": 1.415899987478042e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4aab48c99647d03210aaea8956ce5f9e7e35ae71c20de33f9e619620eb71cb08
-size 111084
+oid sha256:159ee7638163190b38becb35a25396a22ab8d18af56fd258bc432f7e4af2d763
+size 113202