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": 76558.24327816065,
-      "required_pages_per_second_upper_estimate": 0.005296476909581519
+      "required_token_positions_per_second": 76476.26886847423,
+      "required_pages_per_second_upper_estimate": 0.005290805729712578
     },
     {
       "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": 9569.780409770081,
-      "required_pages_per_second_upper_estimate": 0.004006667217831243
+      "required_token_positions_per_second": 9559.533608559279,
+      "required_pages_per_second_upper_estimate": 0.004002377096141822
     },
     {
       "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": 19139.560819540162,
-      "required_pages_per_second_upper_estimate": 0.00398626577454883
+      "required_token_positions_per_second": 19119.067217118558,
+      "required_pages_per_second_upper_estimate": 0.003981997497617049
     },
     {
       "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": 28709.341229310245,
-      "required_pages_per_second_upper_estimate": 0.001721081170071315
+      "required_token_positions_per_second": 28678.600825677837,
+      "required_pages_per_second_upper_estimate": 0.0017192383298114307
     },
     {
       "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": 28709.341229310245,
-      "required_pages_per_second_upper_estimate": 0.002232025001532385
+      "required_token_positions_per_second": 28678.600825677837,
+      "required_pages_per_second_upper_estimate": 0.002229635070362711
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28709.341229310245,
-      "required_pages_per_second_upper_estimate": 0.3682526036006496
+      "required_token_positions_per_second": 28678.600825677837,
+      "required_pages_per_second_upper_estimate": 0.3678582987093269
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1447649596953795,
-    "producer_p95_seconds": 0.31340741900203284,
-    "gpu_update_seconds": 0.9630315018086626,
-    "producer_batches_per_second": 6.907748961518325,
-    "gpu_batches_per_second": 1.038387631268455,
-    "producer_headroom": 6.652379856528222,
+    "producer_batch_seconds": 0.10988336037476643,
+    "producer_p95_seconds": 0.3127588130009826,
+    "gpu_update_seconds": 0.9640637689424837,
+    "producer_batches_per_second": 9.100558961697349,
+    "gpu_batches_per_second": 1.0372757821787413,
+    "producer_headroom": 8.773519172097243,
     "sustainable": true,
-    "buffer_seconds": 60.67098461394574,
+    "buffer_seconds": 60.736017443376475,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 246379.82672048925,
-    "upload_bytes_per_second": 4561559.148643397,
+    "generation_bytes_per_second": 235709.91691588267,
+    "upload_bytes_per_second": 4573220.829429412,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.86737538538941,
+    "estimated_full_shards_per_hour": 50.577861124108885,
     "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": 1.3531316250009695,
-      "bytes": 6148520,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9429252529989753,
       "bytes": 341818,
@@ -562,6 +556,12 @@
       "bytes": 2444431,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.1570196089996898,
+      "bytes": 6331951,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -301,3 +301,4 @@
 {"step": 872600, "loss": 0.5002175569534302, "lr": 3e-05, "grad_norm": 0.2966024002113834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127723944168, "tokens_per_second": 151988.83496253067, "data_wait_s": 1.5334997442550957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872700, "loss": 0.48178231716156006, "lr": 3e-05, "grad_norm": 0.3039153055076488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127738639826, "tokens_per_second": 152296.278396552, "data_wait_s": 1.3289001799421385e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872800, "loss": 0.4873090386390686, "lr": 3e-05, "grad_norm": 0.3048071329696636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127753321336, "tokens_per_second": 152553.46068532305, "data_wait_s": 1.0973999451380223e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 872900, "loss": 0.5579450130462646, "lr": 3e-05, "grad_norm": 0.3107906100421599, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127768125526, "tokens_per_second": 154554.00396586288, "data_wait_s": 8.735001756576821e-06, "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 sha256:7e1fd630d09941cff1ee5189396b009746262bb428fccf211c0308112a3704df
-size 646617
+oid sha256:1a5dac10bc78a7c2d6f403a1d17da4b2d660db9aa2efca015e8c954d47372480
+size 648735