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": 75842.81644751338,
-      "required_pages_per_second_upper_estimate": 0.006573863107809551
+      "required_token_positions_per_second": 74898.93308022647,
+      "required_pages_per_second_upper_estimate": 0.006492049689783646
     },
     {
       "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": 9480.352055939173,
-      "required_pages_per_second_upper_estimate": 0.003781854179008765
+      "required_token_positions_per_second": 9362.36663502831,
+      "required_pages_per_second_upper_estimate": 0.0037347880305681784
     },
     {
       "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": 18960.704111878345,
-      "required_pages_per_second_upper_estimate": 0.004386225199287854
+      "required_token_positions_per_second": 18724.73327005662,
+      "required_pages_per_second_upper_estimate": 0.004331637497977372
     },
     {
       "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": 28441.056167817515,
-      "required_pages_per_second_upper_estimate": 0.001731911101930602
+      "required_token_positions_per_second": 28087.099905084928,
+      "required_pages_per_second_upper_estimate": 0.0017103570225951766
     },
     {
       "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": 28441.056167817515,
-      "required_pages_per_second_upper_estimate": 0.0021894712336810013
+      "required_token_positions_per_second": 28087.099905084928,
+      "required_pages_per_second_upper_estimate": 0.002162222700762201
     },
     {
       "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": 28441.056167817515,
-      "required_pages_per_second_upper_estimate": 0.3531427315124417
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28087.099905084928,
+      "required_pages_per_second_upper_estimate": 0.34874777934471024
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10516362726561113,
-    "producer_p95_seconds": 0.2258753109999816,
-    "gpu_update_seconds": 0.9721157975590619,
-    "producer_batches_per_second": 9.508991140770622,
-    "gpu_batches_per_second": 1.0286840338475665,
-    "producer_headroom": 9.243840506792287,
+    "producer_batch_seconds": 0.2199844752265392,
+    "producer_p95_seconds": 0.215318845001093,
+    "gpu_update_seconds": 0.9843664918568031,
+    "producer_batches_per_second": 4.545775327873495,
+    "gpu_batches_per_second": 1.015881796335537,
+    "producer_headroom": 4.474708912268042,
     "sustainable": true,
-    "buffer_seconds": 60.27117944866183,
+    "buffer_seconds": 60.04635600326499,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321116.82411907864,
-    "upload_bytes_per_second": 5994221.0298787495,
+    "generation_bytes_per_second": 318118.8616019673,
+    "upload_bytes_per_second": 6029943.16648851,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.90419523886938,
+    "estimated_full_shards_per_hour": 68.26090227169288,
     "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": [
@@ -340,6 +340,12 @@
       "bytes": 7704375,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9620766659991205,
+      "bytes": 6846709,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -79,3 +79,4 @@
 {"step": 122900, "loss": 0.6043729484081268, "lr": 0.0003, "grad_norm": 0.13227822076269233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17985221717, "tokens_per_second": 151421.41966565498, "data_wait_s": 1.3359998774831183e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 123000, "loss": 0.6179828643798828, "lr": 0.0003, "grad_norm": 0.13953831270701456, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17999883207, "tokens_per_second": 149125.69639491214, "data_wait_s": 1.533800059405621e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 123100, "loss": 0.53884132578969, "lr": 0.0003, "grad_norm": 0.13073770905537593, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18014480667, "tokens_per_second": 148549.3480506966, "data_wait_s": 1.0952999218716286e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 123200, "loss": 0.6005491511896253, "lr": 0.0003, "grad_norm": 0.12964781555995203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18029136354, "tokens_per_second": 149765.2573419628, "data_wait_s": 1.1881998943863437e-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:e568cea401960c3ba456ce963ee2cf37601c95d7e021956267125230938de49c
-size 172506
+oid sha256:722fa3cd03031d780d3c09411450e8eb5da3e338459f68667c138c173660840c
+size 174624