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": 76094.31899313106,
-      "required_pages_per_second_upper_estimate": 0.005264381551090525
+      "required_token_positions_per_second": 76000.9887537936,
+      "required_pages_per_second_upper_estimate": 0.005257924748577169
     },
     {
       "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": 9511.789874141383,
-      "required_pages_per_second_upper_estimate": 0.003982387791543596
+      "required_token_positions_per_second": 9500.1235942242,
+      "required_pages_per_second_upper_estimate": 0.0039775033637618985
     },
     {
       "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": 19023.579748282766,
-      "required_pages_per_second_upper_estimate": 0.003962109976032447
+      "required_token_positions_per_second": 19000.2471884484,
+      "required_pages_per_second_upper_estimate": 0.00395725041913993
     },
     {
       "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": 28535.36962242415,
-      "required_pages_per_second_upper_estimate": 0.0017106518378777564
+      "required_token_positions_per_second": 28500.3707826726,
+      "required_pages_per_second_upper_estimate": 0.0017085537108747848
     },
     {
       "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": 28535.36962242415,
-      "required_pages_per_second_upper_estimate": 0.002218499474317219
+      "required_token_positions_per_second": 28500.3707826726,
+      "required_pages_per_second_upper_estimate": 0.0022157784684701622
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28535.36962242415,
-      "required_pages_per_second_upper_estimate": 0.3660210826236727
+      "required_token_positions_per_second": 28500.3707826726,
+      "required_pages_per_second_upper_estimate": 0.36557215508616614
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2800141676249268,
-    "producer_p95_seconds": 0.33261145900178235,
-    "gpu_update_seconds": 0.9689028166038957,
-    "producer_batches_per_second": 3.57124787107015,
-    "gpu_batches_per_second": 1.0320952554406881,
-    "producer_headroom": 3.460192121070534,
+    "producer_batch_seconds": 0.09769345686714814,
+    "producer_p95_seconds": 0.3280323799990583,
+    "gpu_update_seconds": 0.9700926423318389,
+    "producer_batches_per_second": 10.23610006307674,
+    "gpu_batches_per_second": 1.030829383053841,
+    "producer_headroom": 9.92996535736322,
     "sustainable": true,
-    "buffer_seconds": 60.07197462944153,
+    "buffer_seconds": 60.14574382457401,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293471.979395355,
-    "upload_bytes_per_second": 2950925.4478334687,
+    "generation_bytes_per_second": 292506.57659787603,
+    "upload_bytes_per_second": 2999984.0722296136,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.97225510020721,
+    "estimated_full_shards_per_hour": 62.76510213329516,
     "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": 2.4877858859981643,
-      "bytes": 7529407,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.8479857339989394,
       "bytes": 6111417,
@@ -562,6 +556,12 @@
       "bytes": 6324939,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.608314076998795,
+      "bytes": 11978117,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -409,3 +409,4 @@
 {"step": 883400, "loss": 0.5266786813735962, "lr": 3e-05, "grad_norm": 0.35947491459576525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129304983932, "tokens_per_second": 125776.63674631277, "data_wait_s": 8.950999472290277e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 883500, "loss": 0.4863302707672119, "lr": 3e-05, "grad_norm": 0.2961368426973886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129319508864, "tokens_per_second": 149855.1608001854, "data_wait_s": 1.3716999092139304e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 883600, "loss": 0.5222635269165039, "lr": 3e-05, "grad_norm": 0.3101059288032435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129334124805, "tokens_per_second": 151940.3420529686, "data_wait_s": 9.590003173798323e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 883700, "loss": 0.5343281030654907, "lr": 3e-05, "grad_norm": 0.3070176355470963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129348785742, "tokens_per_second": 152509.59901306962, "data_wait_s": 1.186799636343494e-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 sha256:5e8f99970a4cb6169344c8e5e666b112c53450bd6b42760ac0e2cba5c0c7e6d5
-size 876927
+oid sha256:4d4356e0135f55d22b02629cbb336353e703c45ca1e7c1c2b73926137e905e11
+size 879045