Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75001.79103747799,
-      "required_pages_per_second_upper_estimate": 0.006500965156827097
+      "required_token_positions_per_second": 74921.68402047131,
+      "required_pages_per_second_upper_estimate": 0.006494021683622337
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9375.223879684749,
-      "required_pages_per_second_upper_estimate": 0.00373991697769457
+      "required_token_positions_per_second": 9365.210502558914,
+      "required_pages_per_second_upper_estimate": 0.003735922491845745
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18750.447759369497,
-      "required_pages_per_second_upper_estimate": 0.004337586092520352
+      "required_token_positions_per_second": 18730.421005117827,
+      "required_pages_per_second_upper_estimate": 0.004332953255383065
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28125.67163905425,
-      "required_pages_per_second_upper_estimate": 0.0017127058388236602
+      "required_token_positions_per_second": 28095.631507676742,
+      "required_pages_per_second_upper_estimate": 0.0017108765524311586
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28125.67163905425,
-      "required_pages_per_second_upper_estimate": 0.002165192059616553
+      "required_token_positions_per_second": 28095.631507676742,
+      "required_pages_per_second_upper_estimate": 0.002162879487146706
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28125.67163905425,
-      "required_pages_per_second_upper_estimate": 0.3492267111893198
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28095.631507676742,
+      "required_pages_per_second_upper_estimate": 0.3488537132954635
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.07106249060959158,
-    "producer_p95_seconds": 0.28519853800025885,
-    "gpu_update_seconds": 0.9830165250741614,
-    "producer_batches_per_second": 14.072121472548362,
-    "gpu_batches_per_second": 1.017276896667182,
-    "producer_headroom": 13.833127950365983,
+    "producer_batch_seconds": 0.12218737527356893,
+    "producer_p95_seconds": 0.3115486620008596,
+    "gpu_update_seconds": 0.9840675762153831,
+    "producer_batches_per_second": 8.184151576716255,
+    "gpu_batches_per_second": 1.0161903757116877,
+    "producer_headroom": 8.053758205478472,
     "sustainable": true,
-    "buffer_seconds": 60.947024554598,
+    "buffer_seconds": 60.02812214913837,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355149.96888912324,
-    "upload_bytes_per_second": 5766665.863937928,
+    "generation_bytes_per_second": 356376.0322782526,
+    "upload_bytes_per_second": 5734460.895811567,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.206915855458,
+    "estimated_full_shards_per_hour": 76.4700005174702,
     "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": 1.2522562880003534,
-      "bytes": 9539459,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 0.8906719839997095,
       "bytes": 2274847,
@@ -562,6 +556,12 @@
       "bytes": 11006785,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.0801306080029462,
+      "bytes": 6178099,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -206,3 +206,4 @@
 {"step": 375600, "loss": 0.5565471760928631, "lr": 0.0003, "grad_norm": 0.15656504648731412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54976547112, "tokens_per_second": 148730.40929264852, "data_wait_s": 1.4162997103994712e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375700, "loss": 0.5926074273884296, "lr": 0.0003, "grad_norm": 0.15002919995749728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54991231745, "tokens_per_second": 149483.97750365784, "data_wait_s": 1.0679999832063913e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 375800, "loss": 0.5588666722178459, "lr": 0.0003, "grad_norm": 0.15751835521936036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55006008728, "tokens_per_second": 149650.65477589727, "data_wait_s": 2.2377000277629122e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 375900, "loss": 0.5704503655433655, "lr": 0.0003, "grad_norm": 0.13811188316819115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55020796021, "tokens_per_second": 151098.66428587525, "data_wait_s": 1.5047000488266349e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:73924395432721072450ee5713c920a4e20b157a3767ab003ca887f4e1c382a5
-size 443841
+oid sha256:19016be3f53a37295b7f8a8c991356afa3fb4e9afdd7fcc5b415eff7e2c2df32
+size 445959