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": 77445.23517208459,
-      "required_pages_per_second_upper_estimate": 0.005357841066907923
+      "required_token_positions_per_second": 75965.7805015586,
+      "required_pages_per_second_upper_estimate": 0.005255488959993151
     },
     {
       "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": 9680.654396510574,
-      "required_pages_per_second_upper_estimate": 0.004053087840767361
+      "required_token_positions_per_second": 9495.722562694826,
+      "required_pages_per_second_upper_estimate": 0.003975660743764539
     },
     {
       "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": 19361.308793021148,
-      "required_pages_per_second_upper_estimate": 0.004032450029537605
+      "required_token_positions_per_second": 18991.44512538965,
+      "required_pages_per_second_upper_estimate": 0.003955417181530805
     },
     {
       "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": 29041.963189531718,
-      "required_pages_per_second_upper_estimate": 0.0017410213487022664
+      "required_token_positions_per_second": 28487.167688084475,
+      "required_pages_per_second_upper_estimate": 0.001707762205514185
     },
     {
       "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": 29041.963189531718,
-      "required_pages_per_second_upper_estimate": 0.0022578848958902213
+      "required_token_positions_per_second": 28487.167688084475,
+      "required_pages_per_second_upper_estimate": 0.0022147519859402115
     },
     {
       "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": 29041.963189531718,
-      "required_pages_per_second_upper_estimate": 0.37251912096473516
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28487.167688084475,
+      "required_pages_per_second_upper_estimate": 0.36540279996516817
     }
   ],
   "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": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09753589247652883,
-    "producer_p95_seconds": 0.18006709400106047,
-    "gpu_update_seconds": 0.9520017575797294,
-    "producer_batches_per_second": 10.252635974399285,
-    "gpu_batches_per_second": 1.0504182287880397,
-    "producer_headroom": 9.76052746745328,
+    "producer_batch_seconds": 0.07393539364836954,
+    "producer_p95_seconds": 0.1946583989993087,
+    "gpu_update_seconds": 0.9705422561739797,
+    "producer_batches_per_second": 13.525321914912839,
+    "gpu_batches_per_second": 1.030351840570185,
+    "producer_headroom": 13.126896446778877,
     "sustainable": true,
-    "buffer_seconds": 60.92811248510268,
+    "buffer_seconds": 60.173619882786745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294756.923208696,
-    "upload_bytes_per_second": 4852924.631568687,
+    "generation_bytes_per_second": 292447.9699346644,
+    "upload_bytes_per_second": 4905477.534129387,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.24797413058911,
+    "estimated_full_shards_per_hour": 62.752526507663234,
     "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": 1.3121002539999154,
-      "bytes": 2619780,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.2438676930000838,
       "bytes": 6205549,
@@ -562,6 +556,12 @@
       "bytes": 12284688,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.182235641999796,
+      "bytes": 6603515,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -93,3 +93,4 @@
 {"step": 851800, "loss": 0.5552166104316711, "lr": 3e-05, "grad_norm": 0.31155580857427795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124679698959, "tokens_per_second": 138299.4433037356, "data_wait_s": 1.0236999514745548e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 851900, "loss": 0.5464333891868591, "lr": 3e-05, "grad_norm": 0.290543992975568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124694460518, "tokens_per_second": 154670.2534661163, "data_wait_s": 1.554200025566388e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852000, "loss": 0.5257346034049988, "lr": 3e-05, "grad_norm": 0.30198060240331964, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124709130689, "tokens_per_second": 144589.06245153135, "data_wait_s": 1.0116998964804225e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 852100, "loss": 0.5304078459739685, "lr": 3e-05, "grad_norm": 0.30842763203355267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124723926557, "tokens_per_second": 153995.97442909091, "data_wait_s": 1.355600034003146e-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 sha256:48632d07179ae6327ddd4c630d3ef2bd98cfc3f8cc1e075d5e60856333304a20
-size 202941
+oid sha256:16a7ebd18244393ed08863b25dd8661529b73fdba9a5b7c37d98bc39d951b6cb
+size 205059