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": 75591.4731367585,
-      "required_pages_per_second_upper_estimate": 0.005229593507983537
+      "required_token_positions_per_second": 75682.86822907437,
+      "required_pages_per_second_upper_estimate": 0.005235916432535779
     },
     {
       "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": 9448.934142094813,
-      "required_pages_per_second_upper_estimate": 0.00395607140911264
+      "required_token_positions_per_second": 9460.358528634297,
+      "required_pages_per_second_upper_estimate": 0.003960854561188403
     },
     {
       "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": 18897.868284189626,
-      "required_pages_per_second_upper_estimate": 0.003935927593296094
+      "required_token_positions_per_second": 18920.717057268594,
+      "required_pages_per_second_upper_estimate": 0.003940686390165776
     },
     {
       "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": 28346.802426284437,
-      "required_pages_per_second_upper_estimate": 0.0016993475223946177
+      "required_token_positions_per_second": 28381.075585902887,
+      "required_pages_per_second_upper_estimate": 0.0017014021459816591
     },
     {
       "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": 28346.802426284437,
-      "required_pages_per_second_upper_estimate": 0.0022038392042367963
+      "required_token_positions_per_second": 28381.075585902887,
+      "required_pages_per_second_upper_estimate": 0.0022065037916454335
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28346.802426284437,
-      "required_pages_per_second_upper_estimate": 0.3636023450992732
+      "required_token_positions_per_second": 28381.075585902887,
+      "required_pages_per_second_upper_estimate": 0.36404196439120695
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09464846033586127,
-    "producer_p95_seconds": 0.3500552620025701,
-    "gpu_update_seconds": 0.9753481039668701,
-    "producer_batches_per_second": 10.56541222595156,
-    "gpu_batches_per_second": 1.0252749720154961,
-    "producer_headroom": 10.304954782210244,
+    "producer_batch_seconds": 0.148743054227225,
+    "producer_p95_seconds": 0.3585896450022119,
+    "gpu_update_seconds": 0.9741702676600807,
+    "producer_batches_per_second": 6.7230030013526925,
+    "gpu_batches_per_second": 1.0265145972910477,
+    "producer_headroom": 6.549349633307279,
     "sustainable": true,
-    "buffer_seconds": 60.47158244594595,
+    "buffer_seconds": 60.398556594925005,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 360572.575836994,
-    "upload_bytes_per_second": 5183673.004013403,
+    "generation_bytes_per_second": 393490.2744715442,
+    "upload_bytes_per_second": 5174347.025328171,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.37048107464184,
+    "estimated_full_shards_per_hour": 84.43385291681047,
     "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": 0.9002165780002542,
-      "bytes": 5078039,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.1330570440004522,
       "bytes": 12010212,
@@ -562,6 +556,12 @@
       "bytes": 6436647,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.4995007219986292,
+      "bytes": 7398571,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -392,3 +392,4 @@
 {"step": 436300, "loss": 0.6405804753303528, "lr": 0.0003, "grad_norm": 0.1689496082436297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63861577180, "tokens_per_second": 150784.97162904218, "data_wait_s": 8.994997187983245e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 436400, "loss": 0.6233225464820862, "lr": 0.0003, "grad_norm": 0.17652776534212125, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63876259565, "tokens_per_second": 151066.72195197208, "data_wait_s": 1.0071998985949904e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436500, "loss": 0.5513224005699158, "lr": 0.0003, "grad_norm": 0.15447092581467953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63890921532, "tokens_per_second": 152774.02123816113, "data_wait_s": 1.5495999832637608e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 436600, "loss": 0.6376541256904602, "lr": 0.0003, "grad_norm": 0.1931321346204208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63905645007, "tokens_per_second": 151648.38823333982, "data_wait_s": 9.438001143280417e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:029fa8d6e98b51e8c124e7f0120dd7ca07dfd46ac7a2de81bf21d21b79f971e6
-size 840921
+oid sha256:47e8a0e712cfb8e6efdae7503cce83ba2a2e097e5756611bc22e9ef491c55e00
+size 843039