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": 77105.20589155932,
-      "required_pages_per_second_upper_estimate": 0.005334317052304555
+      "required_token_positions_per_second": 75098.00970885559,
+      "required_pages_per_second_upper_estimate": 0.005195454563048311
     },
     {
       "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": 9638.150736444915,
-      "required_pages_per_second_upper_estimate": 0.0040352924458752215
+      "required_token_positions_per_second": 9387.251213606949,
+      "required_pages_per_second_upper_estimate": 0.0039302460550407915
     },
     {
       "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": 19276.30147288983,
-      "required_pages_per_second_upper_estimate": 0.004014745246548037
+      "required_token_positions_per_second": 18774.502427213898,
+      "required_pages_per_second_upper_estimate": 0.0039102337386644785
     },
     {
       "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": 28914.452209334748,
-      "required_pages_per_second_upper_estimate": 0.0017333772601374514
+      "required_token_positions_per_second": 28161.75364082085,
+      "required_pages_per_second_upper_estimate": 0.0016882541302592107
     },
     {
       "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": 28914.452209334748,
-      "required_pages_per_second_upper_estimate": 0.002247971478041435
+      "required_token_positions_per_second": 28161.75364082085,
+      "required_pages_per_second_upper_estimate": 0.0021894524751105895
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28914.452209334748,
-      "required_pages_per_second_upper_estimate": 0.3708835470213921
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28161.75364082085,
+      "required_pages_per_second_upper_estimate": 0.3612287379692519
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09159194338303678,
-    "producer_p95_seconds": 0.3191555209996295,
-    "gpu_update_seconds": 0.9562000275790843,
-    "producer_batches_per_second": 10.917990852295905,
-    "gpu_batches_per_second": 1.0458062865066098,
-    "producer_headroom": 10.439783154073535,
+    "producer_batch_seconds": 0.5855742816410157,
+    "producer_p95_seconds": 0.32154535600420786,
+    "gpu_update_seconds": 0.981757043706392,
+    "producer_batches_per_second": 1.7077252730389663,
+    "gpu_batches_per_second": 1.018581945920893,
+    "producer_headroom": 1.6765713155214268,
     "sustainable": true,
-    "buffer_seconds": 60.24060173748231,
+    "buffer_seconds": 60.86893670979631,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 257792.41246344484,
-    "upload_bytes_per_second": 3781779.068014202,
+    "generation_bytes_per_second": 234585.8379765073,
+    "upload_bytes_per_second": 3793164.226317182,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.316250614428604,
+    "estimated_full_shards_per_hour": 50.3366599509374,
     "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": 1.1743250350009475,
-      "bytes": 1256562,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.874451825999131,
       "bytes": 7588108,
@@ -562,6 +556,12 @@
       "bytes": 1293663,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.8874604160009767,
+      "bytes": 5227490,
+      "pages": 13,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -370,3 +370,4 @@
 {"step": 879500, "loss": 0.5753275752067566, "lr": 3e-05, "grad_norm": 0.3235909014660641, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128734278305, "tokens_per_second": 132483.3379214986, "data_wait_s": 1.0771000233944505e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 879600, "loss": 0.5399793386459351, "lr": 3e-05, "grad_norm": 0.3171010540298359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128748742040, "tokens_per_second": 150730.97627141807, "data_wait_s": 1.4767996617592871e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 879700, "loss": 0.4613739848136902, "lr": 3e-05, "grad_norm": 0.30393264649142826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128763382708, "tokens_per_second": 152517.54709122924, "data_wait_s": 1.375400461256504e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 879800, "loss": 0.5388280749320984, "lr": 3e-05, "grad_norm": 0.3172370139995614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128778064251, "tokens_per_second": 123138.25894553463, "data_wait_s": 1.4462995750363916e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec94e7cb8adca2c29b8ab45689bbd3ea01e790ab629c7607a32276497be23604
-size 793542
+oid sha256:dac18e209c7d6668f4d219de6a8f5c50717a87e0ca104eaf97b118ebaf5daaf9
+size 795660