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": 75314.53060852349,
-      "required_pages_per_second_upper_estimate": 0.006528072630214126
+      "required_token_positions_per_second": 75043.04281029209,
+      "required_pages_per_second_upper_estimate": 0.0065045407559429615
     },
     {
       "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": 9414.316326065436,
-      "required_pages_per_second_upper_estimate": 0.003755511539039986
+      "required_token_positions_per_second": 9380.38035128651,
+      "required_pages_per_second_upper_estimate": 0.003741973971312634
     },
     {
       "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": 18828.632652130873,
-      "required_pages_per_second_upper_estimate": 0.00435567279145358
+      "required_token_positions_per_second": 18760.76070257302,
+      "required_pages_per_second_upper_estimate": 0.004339971810428886
     },
     {
       "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": 28242.94897819631,
-      "required_pages_per_second_upper_estimate": 0.0017198474134707666
+      "required_token_positions_per_second": 28141.14105385953,
+      "required_pages_per_second_upper_estimate": 0.0017136478450235546
     },
     {
       "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": 28242.94897819631,
-      "required_pages_per_second_upper_estimate": 0.002174220393828161
+      "required_token_positions_per_second": 28141.14105385953,
+      "required_pages_per_second_upper_estimate": 0.002166382937990344
     },
     {
       "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": 28242.94897819631,
-      "required_pages_per_second_upper_estimate": 0.350682903239459
+      "required_token_positions_per_second": 28141.14105385953,
+      "required_pages_per_second_upper_estimate": 0.3494187895484005
     }
   ],
   "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": 0.1094771668909118,
-    "producer_p95_seconds": 0.3066512090008473,
-    "gpu_update_seconds": 0.9789346013882753,
-    "producer_batches_per_second": 9.134324794835502,
-    "gpu_batches_per_second": 1.0215186985748086,
-    "producer_headroom": 8.941906601983332,
+    "producer_batch_seconds": 0.2968078007657766,
+    "producer_p95_seconds": 0.3005048300001363,
+    "gpu_update_seconds": 0.9824761528711397,
+    "producer_batches_per_second": 3.3691836852668895,
+    "gpu_batches_per_second": 1.0178364096448036,
+    "producer_headroom": 3.3101426254172224,
     "sustainable": true,
-    "buffer_seconds": 60.69394528607307,
+    "buffer_seconds": 60.91352147801066,
     "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": 510034.04297997523,
-    "upload_bytes_per_second": 6103595.5110718,
+    "generation_bytes_per_second": 461394.47120502836,
+    "upload_bytes_per_second": 6066463.313175264,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 109.44143263863985,
+    "estimated_full_shards_per_hour": 99.00451280701769,
     "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": 0.9273372809984721,
-      "bytes": 2566992,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1000417619998188,
       "bytes": 10096800,
@@ -562,6 +556,12 @@
       "bytes": 11754261,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.3580378289989312,
+      "bytes": 2581856,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 97600, "loss": 0.7062435783445835, "lr": 0.0003, "grad_norm": 0.16340439951346944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14282107337, "tokens_per_second": 147795.30343363353, "data_wait_s": 1.7122998542618006e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 97700, "loss": 0.6007490325719118, "lr": 0.0003, "grad_norm": 0.13300135242752106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14296808047, "tokens_per_second": 149448.06676770936, "data_wait_s": 1.540999801363796e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 97800, "loss": 0.6372059918940067, "lr": 0.0003, "grad_norm": 0.14159935106428337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14311376498, "tokens_per_second": 147898.65725491752, "data_wait_s": 1.8279999494552612e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 97900, "loss": 0.6053578369319439, "lr": 0.0003, "grad_norm": 0.13006644836666734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14326096975, "tokens_per_second": 149703.34037703852, "data_wait_s": 1.416500163031742e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0167cc1b66c74ada153d16f8262a33baeca062e29841e7682d0c44119c956c4e
-size 272835
+oid sha256:e44c49f80cf4aa561b26aa38b4fcc326316f4bb2da9184aba3068b09f8aff776
+size 274953