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": 75800.73354959137,
-      "required_pages_per_second_upper_estimate": 0.005244070628894368
+      "required_token_positions_per_second": 77074.41974416704,
+      "required_pages_per_second_upper_estimate": 0.005332187195194253
     },
     {
       "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": 9475.091693698922,
-      "required_pages_per_second_upper_estimate": 0.003967023029737489
+      "required_token_positions_per_second": 9634.30246802088,
+      "required_pages_per_second_upper_estimate": 0.004033681256247061
     },
     {
       "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": 18950.183387397843,
-      "required_pages_per_second_upper_estimate": 0.003946823449652317
+      "required_token_positions_per_second": 19268.60493604176,
+      "required_pages_per_second_upper_estimate": 0.004013142260893911
     },
     {
       "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": 28425.275081096763,
-      "required_pages_per_second_upper_estimate": 0.0017040518382298105
+      "required_token_positions_per_second": 28902.90740406264,
+      "required_pages_per_second_upper_estimate": 0.0017326851666892857
     },
     {
       "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": 28425.275081096763,
-      "required_pages_per_second_upper_estimate": 0.002209940107983724
+      "required_token_positions_per_second": 28902.90740406264,
+      "required_pages_per_second_upper_estimate": 0.0022470739202117606
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28425.275081096763,
-      "required_pages_per_second_upper_estimate": 0.3646089080578336
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28902.90740406264,
+      "required_pages_per_second_upper_estimate": 0.3707354626552076
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.13134523275036258,
-    "producer_p95_seconds": 0.26495692800017423,
-    "gpu_update_seconds": 0.9726554948411507,
-    "producer_batches_per_second": 7.613523376981792,
-    "gpu_batches_per_second": 1.028113248014206,
-    "producer_headroom": 7.405335347722893,
+    "producer_batch_seconds": 0.15351050965634272,
+    "producer_p95_seconds": 0.2608473309992405,
+    "gpu_update_seconds": 0.9565819664257635,
+    "producer_batches_per_second": 6.5142119730997985,
+    "gpu_batches_per_second": 1.0453887226585157,
+    "producer_headroom": 6.231377698942058,
     "sustainable": true,
-    "buffer_seconds": 60.304640680151344,
+    "buffer_seconds": 60.2646638848231,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308504.5019664532,
-    "upload_bytes_per_second": 3596804.0619336134,
+    "generation_bytes_per_second": 313991.71034707635,
+    "upload_bytes_per_second": 3529547.5880852714,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.19788450474927,
+    "estimated_full_shards_per_hour": 67.37531168755739,
     "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": 1.5211821389984834,
-      "bytes": 18268501,
-      "pages": 39,
-      "shards": 5
-    },
     {
       "seconds": 1.3536671069996373,
       "bytes": 6151717,
@@ -562,6 +556,12 @@
       "bytes": 2671177,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.42488349499763,
+      "bytes": 9837963,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -210,3 +210,4 @@
 {"step": 863500, "loss": 0.5279927849769592, "lr": 3e-05, "grad_norm": 0.32024103289731815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126391995489, "tokens_per_second": 152755.33657753884, "data_wait_s": 1.255100141861476e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 863600, "loss": 0.5569524168968201, "lr": 3e-05, "grad_norm": 0.3232156143073694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126406703379, "tokens_per_second": 152431.3429792502, "data_wait_s": 1.4684999769087881e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 863700, "loss": 0.5926230549812317, "lr": 3e-05, "grad_norm": 0.3352663516243666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126421455380, "tokens_per_second": 152880.4483452441, "data_wait_s": 1.38890027301386e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 863800, "loss": 0.5399754643440247, "lr": 3e-05, "grad_norm": 0.3145661950903577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126436236550, "tokens_per_second": 153481.73047923108, "data_wait_s": 1.121400055126287e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04c2504f251f22615c313f0ff3c89ee5b44913040795873f72d34c26f20fe31f
-size 452313
+oid sha256:511d46a9d65f87e2a4b1e40904e62d7e5829961485d6265d68d42569e02e7576
+size 454431