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": 75887.92824925356,
-      "required_pages_per_second_upper_estimate": 0.005250102960536575
+      "required_token_positions_per_second": 73679.84865054431,
+      "required_pages_per_second_upper_estimate": 0.005097342890447327
     },
     {
       "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": 9485.991031156695,
-      "required_pages_per_second_upper_estimate": 0.003971586354727011
+      "required_token_positions_per_second": 9209.98108131804,
+      "required_pages_per_second_upper_estimate": 0.0038560267524727354
     },
     {
       "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": 18971.98206231339,
-      "required_pages_per_second_upper_estimate": 0.003951363538767509
+      "required_token_positions_per_second": 18419.96216263608,
+      "required_pages_per_second_upper_estimate": 0.003836392351408446
     },
     {
       "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": 28457.973093470086,
-      "required_pages_per_second_upper_estimate": 0.001706012034144617
+      "required_token_positions_per_second": 27629.94324395412,
+      "required_pages_per_second_upper_estimate": 0.0016563729089945083
     },
     {
       "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": 28457.973093470086,
-      "required_pages_per_second_upper_estimate": 0.002212482234622393
+      "required_token_positions_per_second": 27629.94324395412,
+      "required_pages_per_second_upper_estimate": 0.002148106555940924
     },
     {
       "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": 28457.973093470086,
-      "required_pages_per_second_upper_estimate": 0.3650283230521682
+      "required_token_positions_per_second": 27629.94324395412,
+      "required_pages_per_second_upper_estimate": 0.35440724521176126
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.11742779875027054,
-    "producer_p95_seconds": 0.2636444810013927,
-    "gpu_update_seconds": 0.9715379204692571,
-    "producer_batches_per_second": 8.515871119466897,
-    "gpu_batches_per_second": 1.0292959018182177,
-    "producer_headroom": 8.273491718391075,
+    "producer_batch_seconds": 0.07578402145304608,
+    "producer_p95_seconds": 0.276183255999058,
+    "gpu_update_seconds": 1.0006535212861805,
+    "producer_batches_per_second": 13.195393709999614,
+    "gpu_batches_per_second": 0.9993469055249609,
+    "producer_headroom": 13.20401718066863,
     "sustainable": true,
-    "buffer_seconds": 60.235351069093944,
+    "buffer_seconds": 60.03921127717083,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331872.2560295619,
+    "generation_bytes_per_second": 331908.3944654053,
     "upload_bytes_per_second": 5139200.681578755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.2120605532183,
+    "estimated_full_shards_per_hour": 71.21981502029057,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 625600, "loss": 0.5008061528205872, "lr": 0.0001, "grad_norm": 0.16755687114194404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91571384588, "tokens_per_second": 153110.7857187314, "data_wait_s": 1.8451002688379958e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625700, "loss": 0.529500424861908, "lr": 0.0001, "grad_norm": 0.1569622526422912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91586040695, "tokens_per_second": 151100.42773856962, "data_wait_s": 1.3044002116657794e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625800, "loss": 0.5401984453201294, "lr": 0.0001, "grad_norm": 0.16631244845804258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91600588830, "tokens_per_second": 150900.90455086165, "data_wait_s": 1.777400029823184e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 625900, "loss": 0.550552487373352, "lr": 0.0001, "grad_norm": 0.1606772406117453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91615241549, "tokens_per_second": 150707.62347697216, "data_wait_s": 1.7790996935218573e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb2febf2e8d6b7a0da307c21b8b8dbc83fcc7cbbf1ab6306b169fb45587fa438
-size 457332
+oid sha256:a713797baaa6f7059974ed1ca13221c0923def0c4ec0255bc3597c59a6439ae6
+size 459450