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": 76013.0212790731,
-      "required_pages_per_second_upper_estimate": 0.006588616031228946
+      "required_token_positions_per_second": 75819.42597709605,
+      "required_pages_per_second_upper_estimate": 0.006571835681116382
     },
     {
       "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": 9501.627659884138,
-      "required_pages_per_second_upper_estimate": 0.003790341335521038
+      "required_token_positions_per_second": 9477.428247137006,
+      "required_pages_per_second_upper_estimate": 0.003780687827962744
     },
     {
       "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": 19003.255319768276,
-      "required_pages_per_second_upper_estimate": 0.004396068672357508
+      "required_token_positions_per_second": 18954.856494274012,
+      "required_pages_per_second_upper_estimate": 0.0043848724558696464
     },
     {
       "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": 28504.88297965241,
-      "required_pages_per_second_upper_estimate": 0.001735797819897946
+      "required_token_positions_per_second": 28432.28474141102,
+      "required_pages_per_second_upper_estimate": 0.0017313769680825642
     },
     {
       "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": 28504.88297965241,
-      "required_pages_per_second_upper_estimate": 0.0021943847983399755
+      "required_token_positions_per_second": 28432.28474141102,
+      "required_pages_per_second_upper_estimate": 0.002188795984995367
     },
     {
       "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": 28504.88297965241,
-      "required_pages_per_second_upper_estimate": 0.3539352469008333
+      "required_token_positions_per_second": 28432.28474141102,
+      "required_pages_per_second_upper_estimate": 0.35303381975254877
     }
   ],
   "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": 0.08891573778106476,
-    "producer_p95_seconds": 0.32703011100238655,
-    "gpu_update_seconds": 0.969939080954513,
-    "producer_batches_per_second": 11.246602963159095,
-    "gpu_batches_per_second": 1.0309925846228447,
-    "producer_headroom": 10.908519741946835,
+    "producer_batch_seconds": 0.22319741781251423,
+    "producer_p95_seconds": 0.37801911600035965,
+    "gpu_update_seconds": 0.9724156975584616,
+    "producer_batches_per_second": 4.4803385711209245,
+    "gpu_batches_per_second": 1.0283667802883036,
+    "producer_headroom": 4.3567515569346345,
     "sustainable": true,
-    "buffer_seconds": 60.1362230191798,
+    "buffer_seconds": 60.28977324862462,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 555445.5537310285,
-    "upload_bytes_per_second": 5111796.448017311,
+    "generation_bytes_per_second": 549008.0145004862,
+    "upload_bytes_per_second": 5070479.625359649,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.18568571994915,
+    "estimated_full_shards_per_hour": 117.80433965931834,
     "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": 1.4903823460008425,
-      "bytes": 7681898,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8942368999996688,
       "bytes": 850966,
@@ -562,6 +556,12 @@
       "bytes": 11040923,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 0.8454459890017461,
+      "bytes": 981466,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -297,3 +297,4 @@
 {"step": 144700, "loss": 0.614732364192605, "lr": 0.0003, "grad_norm": 0.14681944049556186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21176816098, "tokens_per_second": 150433.177234423, "data_wait_s": 1.3577999197877944e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 144800, "loss": 0.6016544653102756, "lr": 0.0003, "grad_norm": 0.132766612257271, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21191490841, "tokens_per_second": 150497.28326587458, "data_wait_s": 9.331000910606235e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 144900, "loss": 0.6100160982459784, "lr": 0.0003, "grad_norm": 0.13386735083109225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21206014750, "tokens_per_second": 148411.26772408662, "data_wait_s": 1.2540996976895258e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 145000, "loss": 0.6024244278669357, "lr": 0.0003, "grad_norm": 0.14424647653323605, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21220526567, "tokens_per_second": 148749.33688451187, "data_wait_s": 1.3396002032095566e-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 sha256:04777791490189bddb13b2cf365bb77534d3f23cf7d83aa5d23b91e29c5b32d7
-size 637362
+oid sha256:40b3d6efc59b312ebc74ea322d56109d49f3e419b8bf04e3695423726413f0ce
+size 639480