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": 75849.31645244265,
-      "required_pages_per_second_upper_estimate": 0.006574426511762709
+      "required_token_positions_per_second": 75009.56664431604,
+      "required_pages_per_second_upper_estimate": 0.006501639126720725
     },
     {
       "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": 9481.164556555332,
-      "required_pages_per_second_upper_estimate": 0.003782178297652518
+      "required_token_positions_per_second": 9376.195830539506,
+      "required_pages_per_second_upper_estimate": 0.003740304703422493
     },
     {
       "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": 18962.329113110663,
-      "required_pages_per_second_upper_estimate": 0.004386601114723905
+      "required_token_positions_per_second": 18752.39166107901,
+      "required_pages_per_second_upper_estimate": 0.0043380357799693466
     },
     {
       "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": 28443.493669665997,
-      "required_pages_per_second_upper_estimate": 0.0017320595330045908
+      "required_token_positions_per_second": 28128.587491618513,
+      "required_pages_per_second_upper_estimate": 0.0017128833989465276
     },
     {
       "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": 28443.493669665997,
-      "required_pages_per_second_upper_estimate": 0.0021896588793207344
+      "required_token_positions_per_second": 28128.587491618513,
+      "required_pages_per_second_upper_estimate": 0.0021654165300185463
     },
     {
       "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": 28443.493669665997,
-      "required_pages_per_second_upper_estimate": 0.3531729971276059
+      "required_token_positions_per_second": 28128.587491618513,
+      "required_pages_per_second_upper_estimate": 0.3492629163194372
     }
   ],
   "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": 0.16153026217176603,
-    "producer_p95_seconds": 0.2692740070015134,
-    "gpu_update_seconds": 0.9720324908429107,
-    "producer_batches_per_second": 6.190790422519296,
-    "gpu_batches_per_second": 1.0287721958067853,
-    "producer_headroom": 6.017649434687867,
+    "producer_batch_seconds": 0.23668176634370752,
+    "producer_p95_seconds": 0.3191470429992478,
+    "gpu_update_seconds": 0.9829146240719796,
+    "producer_batches_per_second": 4.225082546273579,
+    "gpu_batches_per_second": 1.0173823600845817,
+    "producer_headroom": 4.152895422643577,
     "sustainable": true,
-    "buffer_seconds": 60.26601443226046,
+    "buffer_seconds": 60.94070669246273,
     "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": 431100.04687607166,
-    "upload_bytes_per_second": 5747305.471620604,
+    "generation_bytes_per_second": 423684.7406866792,
+    "upload_bytes_per_second": 5636778.74826536,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.504034564129,
+    "estimated_full_shards_per_hour": 90.91288247537882,
     "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": 0.8509849889996985,
-      "bytes": 5837001,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.9184139579992916,
       "bytes": 2547467,
@@ -562,6 +556,12 @@
       "bytes": 5769808,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 2.3888570920025813,
+      "bytes": 6457509,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 134100, "loss": 0.5643828809261322, "lr": 0.0003, "grad_norm": 0.12965959476805836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19625249699, "tokens_per_second": 150435.3461799266, "data_wait_s": 9.978997695725411e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 134200, "loss": 0.6500321365892887, "lr": 0.0003, "grad_norm": 0.15373344914584586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19640017343, "tokens_per_second": 150744.41159973186, "data_wait_s": 1.6715002857381478e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 134300, "loss": 0.6199606731534004, "lr": 0.0003, "grad_norm": 0.1368644183500872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19654668601, "tokens_per_second": 148750.87462226464, "data_wait_s": 1.5558001905446872e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 134400, "loss": 0.5725448876619339, "lr": 0.0003, "grad_norm": 0.1396398416571278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19669231759, "tokens_per_second": 147809.68250423536, "data_wait_s": 1.3234999642008916e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee05e9d1c0d1c4e1abae9bd167dff86c880b7f3e1a1c3ea8ee11123a3e237196
-size 411288
+oid sha256:f7a0c644ae5508d94330a3df514eb0cbaf639a67dfadcfc098d4a13a579ef3c6
+size 413406