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": 75965.4172694068,
-      "required_pages_per_second_upper_estimate": 0.006584489836322331
+      "required_token_positions_per_second": 75843.83563066204,
+      "required_pages_per_second_upper_estimate": 0.006573951448021781
     },
     {
       "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": 9495.67715867585,
-      "required_pages_per_second_upper_estimate": 0.0037879675916211303
+      "required_token_positions_per_second": 9480.479453832755,
+      "required_pages_per_second_upper_estimate": 0.0037819049999332833
     },
     {
       "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": 18991.3543173517,
-      "required_pages_per_second_upper_estimate": 0.004393315584898921
+      "required_token_positions_per_second": 18960.95890766551,
+      "required_pages_per_second_upper_estimate": 0.004386284141808962
     },
     {
       "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": 28487.03147602755,
-      "required_pages_per_second_upper_estimate": 0.0017347107569867931
+      "required_token_positions_per_second": 28441.438361498265,
+      "required_pages_per_second_upper_estimate": 0.0017319343755205444
     },
     {
       "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": 28487.03147602755,
-      "required_pages_per_second_upper_estimate": 0.0021930105401748085
+      "required_token_positions_per_second": 28441.438361498265,
+      "required_pages_per_second_upper_estimate": 0.0021895006560085253
     },
     {
       "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": 28487.03147602755,
-      "required_pages_per_second_upper_estimate": 0.35371359097095184
+      "required_token_positions_per_second": 28441.438361498265,
+      "required_pages_per_second_upper_estimate": 0.35314747707883665
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05751476960938717,
-    "producer_p95_seconds": 0.14836351900021327,
-    "gpu_update_seconds": 0.9705468968666106,
-    "producer_batches_per_second": 17.38683831634069,
-    "gpu_batches_per_second": 1.0303469139188206,
-    "producer_headroom": 16.87474197424594,
+    "producer_batch_seconds": 0.05227423278904553,
+    "producer_p95_seconds": 0.14866290100007973,
+    "gpu_update_seconds": 0.9721027343479098,
+    "producer_batches_per_second": 19.12988381934814,
+    "gpu_batches_per_second": 1.0286978574037278,
+    "producer_headroom": 18.596212368546162,
     "sustainable": true,
-    "buffer_seconds": 60.17390760572985,
+    "buffer_seconds": 60.27036952957041,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -22,3 +22,4 @@
 {"step": 157200, "loss": 0.6457713954150677, "lr": 0.0003, "grad_norm": 0.15602882021255543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23006794010, "tokens_per_second": 149698.9111962909, "data_wait_s": 1.2147000234108418e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157300, "loss": 0.5872071832418442, "lr": 0.0003, "grad_norm": 0.15409774392771908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23021545116, "tokens_per_second": 151733.42046990996, "data_wait_s": 1.871799986474798e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 157400, "loss": 0.6493245474994183, "lr": 0.0003, "grad_norm": 0.15396559682025537, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23036282419, "tokens_per_second": 151325.8073817866, "data_wait_s": 1.5146999885473633e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 157500, "loss": 0.5908354483544827, "lr": 0.0003, "grad_norm": 0.13633144832829183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23051116604, "tokens_per_second": 152658.71742647485, "data_wait_s": 1.6008999864425277e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4d2a12d8994e5ac697eda31137d3038b1073a872dfafeeea4efa210fcdf1fe1c
-size 50997
+oid sha256:b0c7a8a7e8644b370e150eb4870ec14c39d3252e6a2f42e6b764540db879f26a
+size 53115