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": 75641.90267512431,
-      "required_pages_per_second_upper_estimate": 0.006556448411230167
+      "required_token_positions_per_second": 75759.90405085993,
+      "required_pages_per_second_upper_estimate": 0.0065666764714071785
     },
     {
       "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": 9455.237834390538,
-      "required_pages_per_second_upper_estimate": 0.0037718357405419414
+      "required_token_positions_per_second": 9469.988006357491,
+      "required_pages_per_second_upper_estimate": 0.003777719804674282
     },
     {
       "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": 18910.475668781077,
-      "required_pages_per_second_upper_estimate": 0.0043746057329677075
+      "required_token_positions_per_second": 18939.976012714982,
+      "required_pages_per_second_upper_estimate": 0.0043814301183484955
     },
     {
       "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": 28365.713503171613,
-      "required_pages_per_second_upper_estimate": 0.0017273231289460782
+      "required_token_positions_per_second": 28409.964019072475,
+      "required_pages_per_second_upper_estimate": 0.0017300177531999264
     },
     {
       "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": 28365.713503171613,
-      "required_pages_per_second_upper_estimate": 0.0021836711467947156
+      "required_token_positions_per_second": 28409.964019072475,
+      "required_pages_per_second_upper_estimate": 0.002187077674002032
     },
     {
       "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": 28365.713503171613,
-      "required_pages_per_second_upper_estimate": 0.35220722777321745
+      "required_token_positions_per_second": 28409.964019072475,
+      "required_pages_per_second_upper_estimate": 0.35275667108375625
     }
   ],
   "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": 0.09075382133593735,
-    "producer_p95_seconds": 0.20575111099969945,
-    "gpu_update_seconds": 0.9746978512248118,
-    "producer_batches_per_second": 11.018819761851868,
-    "gpu_batches_per_second": 1.0259589664052233,
-    "producer_headroom": 10.740019944910507,
+    "producer_batch_seconds": 0.07192240277342421,
+    "producer_p95_seconds": 0.19523243599905982,
+    "gpu_update_seconds": 0.973179690809853,
+    "producer_batches_per_second": 13.903873639348244,
+    "gpu_batches_per_second": 1.0275594624953877,
+    "producer_headroom": 13.53096744940019,
     "sustainable": true,
-    "buffer_seconds": 60.43126677593833,
+    "buffer_seconds": 60.337140830210885,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277479.04848311335,
-    "upload_bytes_per_second": 4093432.6421139585,
+    "generation_bytes_per_second": 277552.0963886197,
+    "upload_bytes_per_second": 4072833.8578094454,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.54054442281771,
+    "estimated_full_shards_per_hour": 59.55621880287116,
     "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": 0.9509194960000968,
-      "bytes": 6981450,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.187183233000269,
       "bytes": 9995831,
@@ -562,6 +556,12 @@
       "bytes": 11653692,
       "pages": 45,
       "shards": 4
+    },
+    {
+      "seconds": 1.2195578610007942,
+      "bytes": 5944204,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -101,3 +101,4 @@
 {"step": 325100, "loss": 0.6068964377045631, "lr": 0.0003, "grad_norm": 0.1606168216520234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47585915270, "tokens_per_second": 134267.51756507542, "data_wait_s": 1.288900057261344e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325200, "loss": 0.5771619603037834, "lr": 0.0003, "grad_norm": 0.14875200384124426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47600513469, "tokens_per_second": 148641.97463967348, "data_wait_s": 1.4517001545755193e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 325300, "loss": 0.5691273473203182, "lr": 0.0003, "grad_norm": 0.1640405280227942, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47614986024, "tokens_per_second": 147575.77974620298, "data_wait_s": 1.2975000572623685e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 325400, "loss": 0.556862710043788, "lr": 0.0003, "grad_norm": 0.15268652971653798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47629441839, "tokens_per_second": 147466.7865420081, "data_wait_s": 1.2386999515001662e-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:e3b72bf6447d719ba06d25a418b645577976a5dfac8f2e9278f9a620563c29f2
-size 219885
+oid sha256:ae7872fcdf24454aba27628c2aff499bc606b2ce2b7372552e9bd402135da7f2
+size 222003