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": 76530.35858805712,
-      "required_pages_per_second_upper_estimate": 0.005294547782018762
+      "required_token_positions_per_second": 76052.66524384508,
+      "required_pages_per_second_upper_estimate": 0.005261499848064907
     },
     {
       "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": 9566.29482350714,
-      "required_pages_per_second_upper_estimate": 0.004005207875650267
+      "required_token_positions_per_second": 9506.583155480635,
+      "required_pages_per_second_upper_estimate": 0.003980207847168991
     },
     {
       "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": 19132.58964701428,
-      "required_pages_per_second_upper_estimate": 0.003984813863153871
+      "required_token_positions_per_second": 19013.16631096127,
+      "required_pages_per_second_upper_estimate": 0.0039599411316591894
     },
     {
       "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": 28698.884470521418,
-      "required_pages_per_second_upper_estimate": 0.001720454303348471
+      "required_token_positions_per_second": 28519.749466441906,
+      "required_pages_per_second_upper_estimate": 0.001709715433377218
     },
     {
       "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": 28698.884470521418,
-      "required_pages_per_second_upper_estimate": 0.0022312120345309746
+      "required_token_positions_per_second": 28519.749466441906,
+      "required_pages_per_second_upper_estimate": 0.0022172850758954044
     },
     {
       "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": 28698.884470521418,
-      "required_pages_per_second_upper_estimate": 0.3681184755264994
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28519.749466441906,
+      "required_pages_per_second_upper_estimate": 0.3658207240343493
     }
   ],
   "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": 0.09111478200003376,
-    "producer_p95_seconds": 0.26974845999939134,
-    "gpu_update_seconds": 0.9633823930821822,
-    "producer_batches_per_second": 10.975167563915475,
-    "gpu_batches_per_second": 1.0380094209534656,
-    "producer_headroom": 10.573283192202833,
+    "producer_batch_seconds": 0.12213434905484633,
+    "producer_p95_seconds": 0.2939315510011511,
+    "gpu_update_seconds": 0.969433480912318,
+    "producer_batches_per_second": 8.18770483274066,
+    "gpu_batches_per_second": 1.0315302903082286,
+    "producer_headroom": 7.937435196686387,
     "sustainable": true,
-    "buffer_seconds": 60.69309076417748,
+    "buffer_seconds": 60.10487581656371,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437858.3745573505,
-    "upload_bytes_per_second": 3225745.6021400294,
+    "generation_bytes_per_second": 456606.1302953728,
+    "upload_bytes_per_second": 3192815.50960094,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.95421435871492,
+    "estimated_full_shards_per_hour": 97.97704631467712,
     "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": 1.523874344999058,
-      "bytes": 6793396,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.551560215997597,
       "bytes": 6347247,
@@ -562,6 +556,12 @@
       "bytes": 6150813,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.5219117590022506,
+      "bytes": 2396663,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -267,3 +267,4 @@
 {"step": 826700, "loss": 0.5147653222084045, "lr": 3e-05, "grad_norm": 0.27749348607910346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121005498214, "tokens_per_second": 148855.0562910351, "data_wait_s": 1.1134001397294924e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 826800, "loss": 0.5056031942367554, "lr": 3e-05, "grad_norm": 0.2900156346911759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121020137440, "tokens_per_second": 151174.91341448275, "data_wait_s": 8.86499765329063e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 826900, "loss": 0.5027215480804443, "lr": 3e-05, "grad_norm": 0.27999472463592445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121034757006, "tokens_per_second": 151124.35402478158, "data_wait_s": 9.398998372489586e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 827000, "loss": 0.5485939979553223, "lr": 3e-05, "grad_norm": 0.29933487786131624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121049563757, "tokens_per_second": 153425.40087427286, "data_wait_s": 9.273000614484772e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0aa87a47342b74acb492b3c282f95095ffdf6a1a0f0f626cd4e3159a15aef6f1
-size 573822
+oid sha256:c1e6ed56211a5314e3c4c0cc8518696000b61980e39fe8c75f641769d4257449
+size 575940