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": 77336.41633972073,
-      "required_pages_per_second_upper_estimate": 0.0067033245609717665
+      "required_token_positions_per_second": 6324.9279351402065,
+      "required_pages_per_second_upper_estimate": 0.0005482287230346584
     },
     {
       "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": 9667.052042465091,
-      "required_pages_per_second_upper_estimate": 0.003856331595047507
+      "required_token_positions_per_second": 790.6159918925258,
+      "required_pages_per_second_upper_estimate": 0.00031538853992840504
     },
     {
       "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": 19334.104084930183,
-      "required_pages_per_second_upper_estimate": 0.004472604711438319
+      "required_token_positions_per_second": 1581.2319837850516,
+      "required_pages_per_second_upper_estimate": 0.0003657901907162263
     },
     {
       "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": 29001.156127395276,
-      "required_pages_per_second_upper_estimate": 0.001766018250851508
+      "required_token_positions_per_second": 2371.8479756775773,
+      "required_pages_per_second_upper_estimate": 0.0001444330975941687
     },
     {
       "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": 29001.156127395276,
-      "required_pages_per_second_upper_estimate": 0.0022325892790252193
+      "required_token_positions_per_second": 2371.8479756775773,
+      "required_pages_per_second_upper_estimate": 0.00018259142286307977
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 29001.156127395276,
-      "required_pages_per_second_upper_estimate": 0.36009729847641797
+      "estimated_pages_for_buffer": 3,
+      "required_token_positions_per_second": 2371.8479756775773,
+      "required_pages_per_second_upper_estimate": 0.029450413793381643
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,21 +155,21 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 6,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 1.3367408980000164,
-    "producer_p95_seconds": 2.657314213999996,
-    "gpu_update_seconds": 0.9533413040000482,
-    "producer_batches_per_second": 0.7480881309879603,
-    "gpu_batches_per_second": 1.048942278913313,
-    "producer_headroom": 0.7131833143030211,
-    "sustainable": false,
-    "buffer_seconds": 60.06050215200304,
-    "buffer_drain_seconds": 209.40379394613316,
+    "producer_batch_seconds": 0.043788355000002874,
+    "producer_p95_seconds": 0.1254826090000165,
+    "gpu_update_seconds": 11.656733603300044,
+    "producer_batches_per_second": 22.837121878634957,
+    "gpu_batches_per_second": 0.08578732550917163,
+    "producer_headroom": 266.20624600534273,
+    "sustainable": true,
+    "buffer_seconds": 69.94040161980027,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 7,
-    "max_pending_pages": 1780,
+    "upload_pages_per_commit": 23,
+    "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,
     "upload_sustainable": null,

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -0,0 +1 @@
+{"step": 40100, "loss": 0.6127832569181919, "lr": 0.0003, "grad_norm": 0.14283070884692345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5868007171, "tokens_per_second": 46680.67316178548, "data_wait_s": 8.534000016879872e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f930f15e999fb437ffae5ff43bb10a8bef6677821fd507462d8d1f1d5ab87053
-size 165
+oid sha256:f3a49b2f859f7f08ae51667bc1e5cdd9d8c7e17ac9cad05942cb8555728ca557
+size 2283