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": 75807.83965729979,
-      "required_pages_per_second_upper_estimate": 0.006570831408281701
+      "required_token_positions_per_second": 75869.10304161812,
+      "required_pages_per_second_upper_estimate": 0.006576141563163786
     },
     {
       "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": 9475.979957162474,
-      "required_pages_per_second_upper_estimate": 0.003780110083438038
+      "required_token_positions_per_second": 9483.637880202265,
+      "required_pages_per_second_upper_estimate": 0.0037831649434347634
     },
     {
       "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": 18951.959914324947,
-      "required_pages_per_second_upper_estimate": 0.004384202383076496
+      "required_token_positions_per_second": 18967.27576040453,
+      "required_pages_per_second_upper_estimate": 0.0043877454345173225
     },
     {
       "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": 28427.939871487422,
-      "required_pages_per_second_upper_estimate": 0.001731112388299988
+      "required_token_positions_per_second": 28450.913640606796,
+      "required_pages_per_second_upper_estimate": 0.0017325113703053065
     },
     {
       "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": 28427.939871487422,
-      "required_pages_per_second_upper_estimate": 0.0021884615048812788
+      "required_token_positions_per_second": 28450.913640606796,
+      "required_pages_per_second_upper_estimate": 0.0021902300892235513
     },
     {
       "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": 28427.939871487422,
-      "required_pages_per_second_upper_estimate": 0.3529798710094419
+      "required_token_positions_per_second": 28450.913640606796,
+      "required_pages_per_second_upper_estimate": 0.3532651283336454
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.12913467317969207,
-    "producer_p95_seconds": 0.34683890500309644,
-    "gpu_update_seconds": 0.9725643196442214,
-    "producer_batches_per_second": 7.743853570671069,
-    "gpu_batches_per_second": 1.0282096307684976,
-    "producer_headroom": 7.531395679384183,
+    "producer_batch_seconds": 0.08753244829699725,
+    "producer_p95_seconds": 0.3529432600043947,
+    "gpu_update_seconds": 0.9717789857032629,
+    "producer_batches_per_second": 11.424334854738714,
+    "gpu_batches_per_second": 1.0290405685983361,
+    "producer_headroom": 11.101928537472421,
     "sustainable": true,
-    "buffer_seconds": 60.29898781794173,
+    "buffer_seconds": 60.2502971136023,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283180.2533328508,
-    "upload_bytes_per_second": 5268848.120160666,
+    "generation_bytes_per_second": 303003.0014417609,
+    "upload_bytes_per_second": 5375937.692473477,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.76389026631492,
+    "estimated_full_shards_per_hour": 65.017390560528,
     "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": 1.7869264700020722,
-      "bytes": 2542185,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.5899441080000543,
       "bytes": 10502441,
@@ -562,6 +556,12 @@
       "bytes": 16241287,
       "pages": 47,
       "shards": 4
+    },
+    {
+      "seconds": 1.0899500639934558,
+      "bytes": 7493063,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -331,3 +331,4 @@
 {"step": 388100, "loss": 0.6169699393212795, "lr": 0.0003, "grad_norm": 0.15993424197939457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56807507607, "tokens_per_second": 149504.69687023005, "data_wait_s": 1.0605996067170054e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 388200, "loss": 0.5952812060713768, "lr": 0.0003, "grad_norm": 0.16336263963922182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56822193731, "tokens_per_second": 150220.98964047353, "data_wait_s": 1.0478004696778953e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 388300, "loss": 0.5543335117399693, "lr": 0.0003, "grad_norm": 0.16834419330180297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56836712302, "tokens_per_second": 148037.74429061674, "data_wait_s": 1.5964003978297114e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 388400, "loss": 0.5341513473540545, "lr": 0.0003, "grad_norm": 0.12798725855584206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56851283147, "tokens_per_second": 148737.1506964806, "data_wait_s": 1.095100014936179e-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 sha256:8ca10215486bba49484a1e7991bb222d404c7fd586b6f62f14ccc68ddc75ebfb
-size 710157
+oid sha256:a139ec8ad2d7fe2d7a0a4f2b6b78f1f8e7922930e4081a539404ea000a0d8c23
+size 712275