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": 76576.6977968975,
-      "required_pages_per_second_upper_estimate": 0.005297753636008117
+      "required_token_positions_per_second": 75939.38638985518,
+      "required_pages_per_second_upper_estimate": 0.005253662954102732
     },
     {
       "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": 9572.087224612187,
-      "required_pages_per_second_upper_estimate": 0.00400763303303386
+      "required_token_positions_per_second": 9492.423298731897,
+      "required_pages_per_second_upper_estimate": 0.003974279410839727
     },
     {
       "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": 19144.174449224374,
-      "required_pages_per_second_upper_estimate": 0.003987226671942455
+      "required_token_positions_per_second": 18984.846597463795,
+      "required_pages_per_second_upper_estimate": 0.0039540428821787325
     },
     {
       "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": 28716.26167383656,
-      "required_pages_per_second_upper_estimate": 0.0017214960401537606
+      "required_token_positions_per_second": 28477.26989619569,
+      "required_pages_per_second_upper_estimate": 0.0017071688480035053
     },
     {
       "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": 28716.26167383656,
-      "required_pages_per_second_upper_estimate": 0.0022325630356544875
+      "required_token_positions_per_second": 28477.26989619569,
+      "required_pages_per_second_upper_estimate": 0.002213982476156644
     },
     {
       "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": 28716.26167383656,
-      "required_pages_per_second_upper_estimate": 0.3683413716324388
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28477.26989619569,
+      "required_pages_per_second_upper_estimate": 0.36527584171823974
     }
   ],
   "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": 0.09776241564065913,
-    "producer_p95_seconds": 0.27072254399899975,
-    "gpu_update_seconds": 0.96279941707002,
-    "producer_batches_per_second": 10.228879814873382,
-    "gpu_batches_per_second": 1.0386379366983711,
-    "producer_headroom": 9.848359523039386,
+    "producer_batch_seconds": 0.11852821087487087,
+    "producer_p95_seconds": 0.26668486899870913,
+    "gpu_update_seconds": 0.9708795857461577,
+    "producer_batches_per_second": 8.436810043945494,
+    "gpu_batches_per_second": 1.029993847518652,
+    "producer_headroom": 8.191126640484823,
     "sustainable": true,
-    "buffer_seconds": 60.65636327541126,
+    "buffer_seconds": 60.19453431626178,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284667.3512393296,
-    "upload_bytes_per_second": 4821461.536158498,
+    "generation_bytes_per_second": 284595.5912629279,
+    "upload_bytes_per_second": 4886212.894228933,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.082986859177744,
+    "estimated_full_shards_per_hour": 61.067588838728696,
     "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": 1.0786898769983964,
-      "bytes": 798109,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.6809108209999977,
       "bytes": 7259646,
@@ -562,6 +556,12 @@
       "bytes": 6867913,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.275760786000319,
+      "bytes": 7467840,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -267,3 +267,4 @@
 {"step": 869200, "loss": 0.5154692530632019, "lr": 3e-05, "grad_norm": 0.30594704435434894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127226994495, "tokens_per_second": 151496.0445390429, "data_wait_s": 1.036300091072917e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 869300, "loss": 0.5246040225028992, "lr": 3e-05, "grad_norm": 0.3082012577395776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127241499676, "tokens_per_second": 150615.29035726047, "data_wait_s": 9.862997103482485e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 869400, "loss": 0.5129026770591736, "lr": 3e-05, "grad_norm": 0.30051577866975776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127256013327, "tokens_per_second": 150631.43948392197, "data_wait_s": 1.1424999684095383e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 869500, "loss": 0.49472102522850037, "lr": 3e-05, "grad_norm": 0.299259477089786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127270536680, "tokens_per_second": 150904.69734010668, "data_wait_s": 9.16299904929474e-06, "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 sha256:1fd274b8b88a5e42104edfe6fcfb6704478719335b91a9c1f162f200861fd15d
-size 573822
+oid sha256:b56adf98c08b93daa171d95d24ae760bfc3477d3e59882ae4405194b84deb161
+size 575940