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": 77048.32843414017,
-      "required_pages_per_second_upper_estimate": 0.005330382137826414
+      "required_token_positions_per_second": 76413.12669596064,
+      "required_pages_per_second_upper_estimate": 0.005286437407708062
     },
     {
       "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": 9631.041054267522,
-      "required_pages_per_second_upper_estimate": 0.004032315770414594
+      "required_token_positions_per_second": 9551.64083699508,
+      "required_pages_per_second_upper_estimate": 0.003999072557507704
     },
     {
       "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": 19262.082108535044,
-      "required_pages_per_second_upper_estimate": 0.004011783727942791
+      "required_token_positions_per_second": 19103.28167399016,
+      "required_pages_per_second_upper_estimate": 0.003978709785276171
     },
     {
       "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": 28893.123162802563,
-      "required_pages_per_second_upper_estimate": 0.0017320986163654154
+      "required_token_positions_per_second": 28654.92251098524,
+      "required_pages_per_second_upper_estimate": 0.0017178188510003026
     },
     {
       "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": 28893.123162802563,
-      "required_pages_per_second_upper_estimate": 0.0022463132384901185
+      "required_token_positions_per_second": 28654.92251098524,
+      "required_pages_per_second_upper_estimate": 0.002227794185545261
     },
     {
       "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": 28893.123162802563,
-      "required_pages_per_second_upper_estimate": 0.3706099609138231
+      "required_token_positions_per_second": 28654.92251098524,
+      "required_pages_per_second_upper_estimate": 0.36755457871224384
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11676180810175651,
-    "producer_p95_seconds": 0.30260851600178285,
-    "gpu_update_seconds": 0.9569058991723832,
-    "producer_batches_per_second": 8.564444284114819,
-    "gpu_batches_per_second": 1.0450348366175695,
-    "producer_headroom": 8.19536725860267,
+    "producer_batch_seconds": 0.08683022129690698,
+    "producer_p95_seconds": 0.30689670300125726,
+    "gpu_update_seconds": 0.9648604001424459,
+    "producer_batches_per_second": 11.516727529469298,
+    "gpu_batches_per_second": 1.036419361653112,
+    "producer_headroom": 11.11203433241527,
     "sustainable": true,
-    "buffer_seconds": 60.28507164786014,
+    "buffer_seconds": 60.78620520897409,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311799.7088362435,
-    "upload_bytes_per_second": 6008228.040333397,
+    "generation_bytes_per_second": 327857.06613796443,
+    "upload_bytes_per_second": 5745341.256823239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.90495919051627,
+    "estimated_full_shards_per_hour": 70.35049427131844,
     "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": 1.0535939100009273,
-      "bytes": 7640568,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 0.9312373259999731,
       "bytes": 6794347,
@@ -562,6 +556,12 @@
       "bytes": 2441249,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 4.102204398001049,
+      "bytes": 6262944,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -310,3 +310,4 @@
 {"step": 428100, "loss": 0.6150058507919312, "lr": 0.0003, "grad_norm": 0.199323124422195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62661874468, "tokens_per_second": 152834.9155989397, "data_wait_s": 1.4957000530557707e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 428200, "loss": 0.6211259961128235, "lr": 0.0003, "grad_norm": 0.2876302383040058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62676626534, "tokens_per_second": 153195.33449153855, "data_wait_s": 1.528899883851409e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 428300, "loss": 0.5747002959251404, "lr": 0.0003, "grad_norm": 0.1377744653043305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62691255767, "tokens_per_second": 151241.10010620445, "data_wait_s": 1.4224999176803976e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 428400, "loss": 0.5906609296798706, "lr": 0.0003, "grad_norm": 0.16108718480436882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62706002550, "tokens_per_second": 153828.52408089308, "data_wait_s": 9.537998266750947e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8aafa7eb0e2f78014a1687f10b413d2c830d0dd5e54b2931c3cbe2afe75d3c11
-size 665679
+oid sha256:ebbe6eda94236f7c4a130a602bbd4df2e2a02631e4db81720835e0bfe0dcc74c
+size 667797