Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76959.06869686162,
-      "required_pages_per_second_upper_estimate": 0.0053242069420903594
+      "required_token_positions_per_second": 76048.06046628323,
+      "required_pages_per_second_upper_estimate": 0.005261181278868622
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9619.883587107703,
-      "required_pages_per_second_upper_estimate": 0.004027644371909186
+      "required_token_positions_per_second": 9506.007558285404,
+      "required_pages_per_second_upper_estimate": 0.003979966856643183
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19239.767174215405,
-      "required_pages_per_second_upper_estimate": 0.004007136115608402
+      "required_token_positions_per_second": 19012.01511657081,
+      "required_pages_per_second_upper_estimate": 0.003959701368226693
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28859.650761323108,
-      "required_pages_per_second_upper_estimate": 0.0017300919970061198
+      "required_token_positions_per_second": 28518.022674856216,
+      "required_pages_per_second_upper_estimate": 0.0017096119148583139
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28859.650761323108,
-      "required_pages_per_second_upper_estimate": 0.002243710906504625
+      "required_token_positions_per_second": 28518.022674856216,
+      "required_pages_per_second_upper_estimate": 0.0022171508254449704
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28859.650761323108,
-      "required_pages_per_second_upper_estimate": 0.37018061288750925
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28518.022674856216,
+      "required_pages_per_second_upper_estimate": 0.3657985746059724
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08294348313280686,
-    "producer_p95_seconds": 0.14168488799987244,
-    "gpu_update_seconds": 0.9580157510794646,
-    "producer_batches_per_second": 12.056402290205574,
-    "gpu_batches_per_second": 1.0438241739483185,
-    "producer_headroom": 11.55022329536747,
+    "producer_batch_seconds": 0.2953372195703068,
+    "producer_p95_seconds": 0.14303036199999042,
+    "gpu_update_seconds": 0.9694921809700608,
+    "producer_batches_per_second": 3.385959959448809,
+    "gpu_batches_per_second": 1.0314678340153434,
+    "producer_headroom": 3.2826617057633247,
     "sustainable": true,
-    "buffer_seconds": 60.35499231800627,
+    "buffer_seconds": 60.10851522014377,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 69,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330585.75170001294,
-    "upload_bytes_per_second": 4427465.34486943,
+    "generation_bytes_per_second": 314159.5747552844,
+    "upload_bytes_per_second": 4633225.538726667,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.93600667238513,
+    "estimated_full_shards_per_hour": 67.41133148187541,
     "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": [
@@ -256,6 +256,12 @@
       "bytes": 1132423,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 0.8866051049999442,
+      "bytes": 7112636,
+      "pages": 23,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -13,3 +13,4 @@
 {"step": 398400, "loss": 0.6168960332870483, "lr": 0.0003, "grad_norm": 0.16448830891151384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58314551555, "tokens_per_second": 148816.51930657792, "data_wait_s": 7.1260001277551055e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 398500, "loss": 0.5579304099082947, "lr": 0.0003, "grad_norm": 0.13817116835845394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58329052213, "tokens_per_second": 151002.1608199904, "data_wait_s": 9.876000149233732e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 398600, "loss": 0.5956971049308777, "lr": 0.0003, "grad_norm": 0.14707166224125492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58343660275, "tokens_per_second": 151360.93894249553, "data_wait_s": 8.527000090907677e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 398700, "loss": 0.6194034218788147, "lr": 0.0003, "grad_norm": 0.14603193163626443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58358359562, "tokens_per_second": 152771.0292765205, "data_wait_s": 9.671000043454114e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b5196afd6a62b6852fcd77b80b638f1f5926f1988b80a47a015d6ffa44a9019
-size 31935
+oid sha256:60261d07e3e08fea6e4bf6cb26064f528badc1bf53de5255478a1bbc931a4e22
+size 34053