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": 75901.6656764869,
-      "required_pages_per_second_upper_estimate": 0.006578964009824969
+      "required_token_positions_per_second": 75738.1373972849,
+      "required_pages_per_second_upper_estimate": 0.006564789793042374
     },
     {
       "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": 9487.708209560862,
-      "required_pages_per_second_upper_estimate": 0.003784788658672755
+      "required_token_positions_per_second": 9467.267174660612,
+      "required_pages_per_second_upper_estimate": 0.0037766344242303384
     },
     {
       "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": 18975.416419121724,
-      "required_pages_per_second_upper_estimate": 0.0043896286326408455
+      "required_token_positions_per_second": 18934.534349321224,
+      "required_pages_per_second_upper_estimate": 0.004380171285292354
     },
     {
       "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": 28463.124628682584,
-      "required_pages_per_second_upper_estimate": 0.001733254955413019
+      "required_token_positions_per_second": 28401.801523981838,
+      "required_pages_per_second_upper_estimate": 0.0017295207000742087
     },
     {
       "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": 28463.124628682584,
-      "required_pages_per_second_upper_estimate": 0.002191170124887804
+      "required_token_positions_per_second": 28401.801523981838,
+      "required_pages_per_second_upper_estimate": 0.002186449302534722
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28463.124628682584,
-      "required_pages_per_second_upper_estimate": 0.3534167479379985
+      "required_token_positions_per_second": 28401.801523981838,
+      "required_pages_per_second_upper_estimate": 0.35265532021284424
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.14093382614831285,
-    "producer_p95_seconds": 0.35941645599814365,
-    "gpu_update_seconds": 0.9713620820160702,
-    "producer_batches_per_second": 7.095528641560061,
-    "gpu_batches_per_second": 1.0294822275999198,
-    "producer_headroom": 6.892327474270439,
+    "producer_batch_seconds": 0.09441961596070314,
+    "producer_p95_seconds": 0.3767981029996008,
+    "gpu_update_seconds": 0.9734593763939465,
+    "producer_batches_per_second": 10.591019565427949,
+    "gpu_batches_per_second": 1.0272642333616115,
+    "producer_headroom": 10.309927301537577,
     "sustainable": true,
-    "buffer_seconds": 60.22444908499635,
+    "buffer_seconds": 60.35448133642468,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436453.6801078162,
-    "upload_bytes_per_second": 3147383.6540707,
+    "generation_bytes_per_second": 367606.71777877514,
+    "upload_bytes_per_second": 3175017.952221257,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.65279962945809,
+    "estimated_full_shards_per_hour": 78.8798441889042,
     "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": 1.1586719139995694,
-      "bytes": 7299186,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 3.9096268039975257,
       "bytes": 6182999,
@@ -562,6 +556,12 @@
       "bytes": 2606249,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0264649749988166,
+      "bytes": 10585305,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -315,3 +315,4 @@
 {"step": 346500, "loss": 0.622475216165185, "lr": 0.0003, "grad_norm": 0.1781314636102928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50718501158, "tokens_per_second": 148319.9760874813, "data_wait_s": 1.1948999599553645e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346600, "loss": 0.595458822324872, "lr": 0.0003, "grad_norm": 0.15219270894335488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50733088387, "tokens_per_second": 148684.9331796579, "data_wait_s": 1.6601999959675595e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346700, "loss": 0.6234288327395916, "lr": 0.0003, "grad_norm": 0.1694497640742114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50747791494, "tokens_per_second": 138618.26169335644, "data_wait_s": 1.367900040349923e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 346800, "loss": 0.5871155746281147, "lr": 0.0003, "grad_norm": 0.14986730547354005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50762460943, "tokens_per_second": 148582.02651873155, "data_wait_s": 1.7239999579032883e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:04d16d81669b48c6f9f5572a95ec5642622068752a157f6bf5aa5ed4a1fc97d0
-size 676269
+oid sha256:f1708e02dfa3aacf57e1e4dc634d0e1037a7d071dc9dd7b686e6f51ed9e8663d
+size 678387