Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 78510.47024088624,
-      "required_pages_per_second_upper_estimate": 0.005431536500653512
+      "required_token_positions_per_second": 78101.12507902418,
+      "required_pages_per_second_upper_estimate": 0.0054032170525442616
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9813.80878011078,
-      "required_pages_per_second_upper_estimate": 0.0041088368006010475
+      "required_token_positions_per_second": 9762.640634878022,
+      "required_pages_per_second_upper_estimate": 0.004087413766704469
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19627.61756022156,
-      "required_pages_per_second_upper_estimate": 0.004087915122710981
+      "required_token_positions_per_second": 19525.281269756044,
+      "required_pages_per_second_upper_estimate": 0.00406660117219648
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29441.42634033234,
-      "required_pages_per_second_upper_estimate": 0.0017649685546478494
+      "required_token_positions_per_second": 29287.92190463407,
+      "required_pages_per_second_upper_estimate": 0.00175576619811544
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29441.42634033234,
-      "required_pages_per_second_upper_estimate": 0.00228894139881226
+      "required_token_positions_per_second": 29287.92190463407,
+      "required_pages_per_second_upper_estimate": 0.0022770071041313756
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29441.42634033234,
-      "required_pages_per_second_upper_estimate": 0.37764300535309114
+      "required_token_positions_per_second": 29287.92190463407,
+      "required_pages_per_second_upper_estimate": 0.3756740152721754
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,27 +158,40 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 1.2944492979999893,
-    "producer_p95_seconds": 2.576500048000014,
-    "gpu_update_seconds": 0.9390849369999614,
-    "producer_batches_per_second": 0.7725292922210757,
-    "gpu_batches_per_second": 1.0648664040918816,
-    "producer_headroom": 0.7254706217160536,
-    "sustainable": false,
-    "buffer_seconds": 60.10143596799753,
-    "buffer_drain_seconds": 218.92533448946386,
+    "producer_batch_seconds": 0.04390004595312824,
+    "producer_p95_seconds": 0.11876418500003183,
+    "gpu_update_seconds": 0.9440068875499633,
+    "producer_batches_per_second": 22.77901943582685,
+    "gpu_batches_per_second": 1.0593143049997855,
+    "producer_headroom": 21.503551239055025,
+    "sustainable": true,
+    "buffer_seconds": 60.41644080319765,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 6,
-    "max_pending_pages": 1465,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "upload_pages_per_commit": 17,
+    "max_pending_pages": 4096,
+    "generation_bytes_per_second": 282769.1575946625,
+    "upload_bytes_per_second": 47836.798533638605,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 60.67567869071871,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.4923919180000098,
+      "bytes": 40304,
+      "pages": 1,
+      "shards": 1
+    },
+    {
+      "seconds": 1.3177459699999758,
+      "bytes": 94124,
+      "pages": 4,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -0,0 +1 @@
+{"step": 842600, "loss": 0.5411972999572754, "lr": 3e-05, "grad_norm": 0.28530642342954515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123332630293, "tokens_per_second": 152128.42886191164, "data_wait_s": 1.4998999972704041e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f55a3d6927594e3f94c23b592fcd9076e21190d6cbbfb902c83fc265172973f
-size 165
+oid sha256:988b04477432c76bb504d887cd4e79fa02bb8eeba90f74837a38eda19991c5c5
+size 2283