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": 74951.52770393826,
-      "required_pages_per_second_upper_estimate": 0.0051853205980630515
+      "required_token_positions_per_second": 76764.1500635525,
+      "required_pages_per_second_upper_estimate": 0.005310722018764494
     },
     {
       "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": 9368.940962992283,
-      "required_pages_per_second_upper_estimate": 0.0039225799354699435
+      "required_token_positions_per_second": 9595.518757944063,
+      "required_pages_per_second_upper_estimate": 0.00401744332673386
     },
     {
       "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": 18737.881925984566,
-      "required_pages_per_second_upper_estimate": 0.0039026066540059695
+      "required_token_positions_per_second": 19191.037515888125,
+      "required_pages_per_second_upper_estimate": 0.003996987012866338
     },
     {
       "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": 28106.82288897685,
-      "required_pages_per_second_upper_estimate": 0.0016849611155605644
+      "required_token_positions_per_second": 28786.556273832186,
+      "required_pages_per_second_upper_estimate": 0.001725710094089859
     },
     {
       "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": 28106.82288897685,
-      "required_pages_per_second_upper_estimate": 0.002185181850769561
+      "required_token_positions_per_second": 28786.556273832186,
+      "required_pages_per_second_upper_estimate": 0.0022380281316108784
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28106.82288897685,
-      "required_pages_per_second_upper_estimate": 0.3605241452646432
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28786.556273832186,
+      "required_pages_per_second_upper_estimate": 0.3692430352847217
     }
   ],
   "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": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3977010485546515,
-    "producer_p95_seconds": 0.22356627399858553,
-    "gpu_update_seconds": 0.9836757469604722,
-    "producer_batches_per_second": 2.5144515048030645,
-    "gpu_batches_per_second": 1.0165951565746836,
-    "producer_headroom": 2.4734049621830376,
+    "producer_batch_seconds": 0.05465039251568271,
+    "producer_p95_seconds": 0.18919530299899634,
+    "gpu_update_seconds": 0.960448333485893,
+    "producer_batches_per_second": 18.29813024148062,
+    "gpu_batches_per_second": 1.0411804207838609,
+    "producer_headroom": 17.574408696337883,
     "sustainable": true,
-    "buffer_seconds": 60.004220564588806,
+    "buffer_seconds": 60.50824500961126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311573.3286201424,
-    "upload_bytes_per_second": 5521170.65774714,
+    "generation_bytes_per_second": 311488.6517984782,
+    "upload_bytes_per_second": 5561768.037703326,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.8563832660027,
+    "estimated_full_shards_per_hour": 66.83821359124909,
     "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": 1.0429953859998022,
-      "bytes": 2436103,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.247582260999934,
       "bytes": 10464030,
@@ -562,6 +556,12 @@
       "bytes": 5785267,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.081288327000948,
+      "bytes": 5787210,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 614400, "loss": 0.5308977961540222, "lr": 0.0001, "grad_norm": 0.15364842307200952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89930116871, "tokens_per_second": 151912.3556869722, "data_wait_s": 1.3777998901787214e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614500, "loss": 0.6621916890144348, "lr": 0.0001, "grad_norm": 0.1962394419839126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89944810027, "tokens_per_second": 152048.15024245056, "data_wait_s": 1.0489000487723388e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614600, "loss": 0.5378708839416504, "lr": 0.0001, "grad_norm": 0.1514035896493702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89959443471, "tokens_per_second": 138751.9963209251, "data_wait_s": 1.4707000445923768e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 614700, "loss": 0.4195510745048523, "lr": 0.0001, "grad_norm": 0.20096422566688188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89974139719, "tokens_per_second": 153044.26359626112, "data_wait_s": 1.243699989572633e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ea34db27acafd7fda98374ed3c045886275baec8aeca54f0d386fb235f47097c
-size 217767
+oid sha256:f6ada174c40bb85479ad9ba1a865052a808770c6020e44c01dc1f90e732efb89
+size 219885