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": 75028.09166161777,
-      "required_pages_per_second_upper_estimate": 0.006503244828269211
+      "required_token_positions_per_second": 74936.19790478345,
+      "required_pages_per_second_upper_estimate": 0.0064952797103294095
     },
     {
       "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": 9378.511457702221,
-      "required_pages_per_second_upper_estimate": 0.0037412284417194116
+      "required_token_positions_per_second": 9367.02473809793,
+      "required_pages_per_second_upper_estimate": 0.0037366462175274975
     },
     {
       "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": 18757.022915404443,
-      "required_pages_per_second_upper_estimate": 0.004339107139150243
+      "required_token_positions_per_second": 18734.04947619586,
+      "required_pages_per_second_upper_estimate": 0.004333792638308059
     },
     {
       "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": 28135.534373106664,
-      "required_pages_per_second_upper_estimate": 0.0017133064275816857
+      "required_token_positions_per_second": 28101.074214293792,
+      "required_pages_per_second_upper_estimate": 0.0017112079847084623
     },
     {
       "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": 28135.534373106664,
-      "required_pages_per_second_upper_estimate": 0.0021659513201857028
+      "required_token_positions_per_second": 28101.074214293792,
+      "required_pages_per_second_upper_estimate": 0.002163298481768463
     },
     {
       "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": 28135.534373106664,
-      "required_pages_per_second_upper_estimate": 0.3493491733378033
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28101.074214293792,
+      "required_pages_per_second_upper_estimate": 0.34892129349608
     }
   ],
   "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": 0.16803091498434242,
-    "producer_p95_seconds": 0.19147145900024043,
-    "gpu_update_seconds": 0.9826719348336717,
-    "producer_batches_per_second": 5.951285810073597,
-    "gpu_batches_per_second": 1.0176336217124806,
-    "producer_headroom": 5.848161541733197,
+    "producer_batch_seconds": 0.07818848157035063,
+    "producer_p95_seconds": 0.21228436499950476,
+    "gpu_update_seconds": 0.9838769788358007,
+    "producer_batches_per_second": 12.78960762398542,
+    "gpu_batches_per_second": 1.016387232866529,
+    "producer_headroom": 12.583400509582098,
     "sustainable": true,
-    "buffer_seconds": 60.92565995968764,
+    "buffer_seconds": 60.01649570898385,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 467484.8097554257,
-    "upload_bytes_per_second": 5988381.0431731995,
+    "generation_bytes_per_second": 468427.2937783392,
+    "upload_bytes_per_second": 5963894.332241848,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.31135768410758,
+    "estimated_full_shards_per_hour": 100.51359281551964,
     "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": [
@@ -400,6 +400,12 @@
       "bytes": 6163490,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.1912384690003819,
+      "bytes": 6120903,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -92,3 +92,4 @@
 {"step": 124200, "loss": 0.6450304668396711, "lr": 0.0003, "grad_norm": 0.1337577137552498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18175258782, "tokens_per_second": 151434.2018835993, "data_wait_s": 1.2652000805246644e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 124300, "loss": 0.6086334884166718, "lr": 0.0003, "grad_norm": 0.1368946043999693, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18190034351, "tokens_per_second": 151453.04840488223, "data_wait_s": 9.346998922410421e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 124400, "loss": 0.619434978812933, "lr": 0.0003, "grad_norm": 0.14032894751096783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18204768422, "tokens_per_second": 150585.28200775647, "data_wait_s": 1.2665001122513786e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 124500, "loss": 0.5670794546604156, "lr": 0.0003, "grad_norm": 0.13992898670251436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18219316192, "tokens_per_second": 149352.98389062865, "data_wait_s": 8.661998435854912e-06, "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:ab493dfb02cffa367fd3642538514c67a859bc70bb86d8e267ab978db911c08a
-size 200040
+oid sha256:019aeeff923734335cceac575c66eda6bb5caafa60147bd8f5c6dcac2cb5c50d
+size 202158