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": 77185.15560914169,
-      "required_pages_per_second_upper_estimate": 0.0053398481592758056
+      "required_token_positions_per_second": 75922.44010567981,
+      "required_pages_per_second_upper_estimate": 0.0052524905708426824
     },
     {
       "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": 9648.14445114271,
-      "required_pages_per_second_upper_estimate": 0.004039476605526694
+      "required_token_positions_per_second": 9490.305013209976,
+      "required_pages_per_second_upper_estimate": 0.00397339252892653
     },
     {
       "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": 19296.28890228542,
-      "required_pages_per_second_upper_estimate": 0.004018908100987179
+      "required_token_positions_per_second": 18980.610026419952,
+      "required_pages_per_second_upper_estimate": 0.003953160516156192
     },
     {
       "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": 28944.433353428132,
-      "required_pages_per_second_upper_estimate": 0.0017351745839472936
+      "required_token_positions_per_second": 28470.91503962993,
+      "required_pages_per_second_upper_estimate": 0.0017067878840556923
     },
     {
       "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": 28944.433353428132,
-      "required_pages_per_second_upper_estimate": 0.0022503023801215796
+      "required_token_positions_per_second": 28470.91503962993,
+      "required_pages_per_second_upper_estimate": 0.002213488414010716
     },
     {
       "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": 28944.433353428132,
-      "required_pages_per_second_upper_estimate": 0.3712681129465775
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28470.91503962993,
+      "required_pages_per_second_upper_estimate": 0.36519432844152755
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.35661916875011457,
-    "producer_p95_seconds": 0.2514802379992034,
-    "gpu_update_seconds": 0.955209579071805,
-    "producer_batches_per_second": 2.8041117461655762,
-    "gpu_batches_per_second": 1.046890673952117,
-    "producer_headroom": 2.678514400725124,
+    "producer_batch_seconds": 0.09581835446078912,
+    "producer_p95_seconds": 0.24490090900144423,
+    "gpu_update_seconds": 0.9710962911278237,
+    "producer_batches_per_second": 10.436413833523106,
+    "gpu_batches_per_second": 1.0297639988292075,
+    "producer_headroom": 10.1347627664094,
     "sustainable": true,
-    "buffer_seconds": 60.17820348152372,
+    "buffer_seconds": 60.207970049925066,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300136.9679538637,
-    "upload_bytes_per_second": 5438002.823820035,
+    "generation_bytes_per_second": 301684.1924153901,
+    "upload_bytes_per_second": 5493930.116593898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.40240649187025,
+    "estimated_full_shards_per_hour": 64.73440484377171,
     "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": 1.0752210870014096,
-      "bytes": 5610482,
-      "pages": 19,
-      "shards": 2
-    },
     {
       "seconds": 1.425927022999531,
       "bytes": 10079783,
@@ -562,6 +556,12 @@
       "bytes": 13168093,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 1.1610033780016238,
+      "bytes": 10490094,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -246,3 +246,4 @@
 {"step": 581700, "loss": 0.578392505645752, "lr": 0.0003, "grad_norm": 0.17558227191054976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85144395957, "tokens_per_second": 154703.03630238463, "data_wait_s": 1.0376999853178859e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 581800, "loss": 0.5910410284996033, "lr": 0.0003, "grad_norm": 0.19374622682089612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85158879530, "tokens_per_second": 150689.796300987, "data_wait_s": 8.909999451134354e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 581900, "loss": 0.5755289793014526, "lr": 0.0003, "grad_norm": 0.15386969171978265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85173342640, "tokens_per_second": 150449.20960974216, "data_wait_s": 9.972998668672517e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 582000, "loss": 0.56739342212677, "lr": 0.0003, "grad_norm": 0.163571416364174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85187903483, "tokens_per_second": 151341.70474212623, "data_wait_s": 1.1173000530106947e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:01ef0898097c47d2af638c35e1266c875f7d719e9553a7a99f9f305173e64d8c
-size 529344
+oid sha256:12bea21c2a42d26f5074d7c53a41296d70adef5e32e029bc580e24b30a561f7b
+size 531462