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": 75073.04487593775,
-      "required_pages_per_second_upper_estimate": 0.005193727437980661
+      "required_token_positions_per_second": 77235.94945110014,
+      "required_pages_per_second_upper_estimate": 0.005343362195120442
     },
     {
       "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": 9384.130609492218,
-      "required_pages_per_second_upper_estimate": 0.003928939523263578
+      "required_token_positions_per_second": 9654.493681387517,
+      "required_pages_per_second_upper_estimate": 0.004042134895643191
     },
     {
       "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": 18768.261218984437,
-      "required_pages_per_second_upper_estimate": 0.0039089338595820105
+      "required_token_positions_per_second": 19308.987362775035,
+      "required_pages_per_second_upper_estimate": 0.004021552855426243
     },
     {
       "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": 28152.391828476655,
-      "required_pages_per_second_upper_estimate": 0.0016876929039038359
+      "required_token_positions_per_second": 28963.481044162552,
+      "required_pages_per_second_upper_estimate": 0.0017363164639226823
     },
     {
       "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": 28152.391828476655,
-      "required_pages_per_second_upper_estimate": 0.0021887246353790887
+      "required_token_positions_per_second": 28963.481044162552,
+      "required_pages_per_second_upper_estimate": 0.0022517832542942435
     },
     {
       "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": 28152.391828476655,
-      "required_pages_per_second_upper_estimate": 0.36110865469243153
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28963.481044162552,
+      "required_pages_per_second_upper_estimate": 0.3715124362715018
     }
   ],
   "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": 0.10718678771894474,
-    "producer_p95_seconds": 0.30629233900253894,
-    "gpu_update_seconds": 0.9820835177504722,
-    "producer_batches_per_second": 9.329508060472035,
-    "gpu_batches_per_second": 1.0182433387035827,
-    "producer_headroom": 9.162356094909763,
+    "producer_batch_seconds": 0.11561486713293334,
+    "producer_p95_seconds": 0.31231984800251666,
+    "gpu_update_seconds": 0.9545813901941983,
+    "producer_batches_per_second": 8.649406644650687,
+    "gpu_batches_per_second": 1.0475796095255554,
+    "producer_headroom": 8.256562619205589,
     "sustainable": true,
-    "buffer_seconds": 60.88917810052928,
+    "buffer_seconds": 60.13862758223449,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 226570.51891569054,
-    "upload_bytes_per_second": 4649991.6172501845,
+    "generation_bytes_per_second": 227333.44084289653,
+    "upload_bytes_per_second": 4681275.070577774,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.61675906756437,
+    "estimated_full_shards_per_hour": 48.780464353229256,
     "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": 1.3247330819976924,
-      "bytes": 862254,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.7279805650032358,
       "bytes": 13248087,
@@ -562,6 +556,12 @@
       "bytes": 6228251,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0129257990010956,
+      "bytes": 2288801,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -306,3 +306,4 @@
 {"step": 873100, "loss": 0.5723668932914734, "lr": 3e-05, "grad_norm": 0.42357364625010396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127797571780, "tokens_per_second": 152783.4733376409, "data_wait_s": 9.731000318424776e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 873200, "loss": 0.49504029750823975, "lr": 3e-05, "grad_norm": 0.4155996004658534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127812119291, "tokens_per_second": 151725.32101300327, "data_wait_s": 1.5221001376630738e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 873300, "loss": 0.5646331906318665, "lr": 3e-05, "grad_norm": 0.32176437313481454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127826608135, "tokens_per_second": 150051.26434386065, "data_wait_s": 1.41079981403891e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 873400, "loss": 0.5732330083847046, "lr": 3e-05, "grad_norm": 0.319052623109754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127841244868, "tokens_per_second": 151732.2769361672, "data_wait_s": 1.0592000762699172e-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:fb471686de38eb5407d475e3764c4bd99857dfdcd84a35f3cf00c89c54e9ade3
-size 657207
+oid sha256:e8c92a7c2c9ea4c0e8769fd052bf50381cfdd6cfcd05943312af8919754bad55
+size 659325