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": 75650.85152719976,
-      "required_pages_per_second_upper_estimate": 0.006557224075047409
+      "required_token_positions_per_second": 74706.8448891188,
+      "required_pages_per_second_upper_estimate": 0.006475399972221502
     },
     {
       "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": 9456.35644089997,
-      "required_pages_per_second_upper_estimate": 0.003772281969403211
+      "required_token_positions_per_second": 9338.35561113985,
+      "required_pages_per_second_upper_estimate": 0.003725209674142273
     },
     {
       "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": 18912.71288179994,
-      "required_pages_per_second_upper_estimate": 0.004375123272826017
+      "required_token_positions_per_second": 18676.7112222797,
+      "required_pages_per_second_upper_estimate": 0.004320528442383358
     },
     {
       "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": 28369.069322699914,
-      "required_pages_per_second_upper_estimate": 0.0017275274807487275
+      "required_token_positions_per_second": 28015.066833419554,
+      "required_pages_per_second_upper_estimate": 0.0017059705864590745
     },
     {
       "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": 28369.069322699914,
-      "required_pages_per_second_upper_estimate": 0.002183929487071507
+      "required_token_positions_per_second": 28015.066833419554,
+      "required_pages_per_second_upper_estimate": 0.0021566773955050958
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28369.069322699914,
-      "required_pages_per_second_upper_estimate": 0.35224889582055346
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28015.066833419554,
+      "required_pages_per_second_upper_estimate": 0.347853369673809
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.24089868162499073,
-    "producer_p95_seconds": 0.17457737699987774,
-    "gpu_update_seconds": 0.9745825527620344,
-    "producer_batches_per_second": 4.151122759387739,
-    "gpu_batches_per_second": 1.0260803429795975,
-    "producer_headroom": 4.045611815672683,
+    "producer_batch_seconds": 0.05229192767972535,
+    "producer_p95_seconds": 0.16628922899963072,
+    "gpu_update_seconds": 0.9868975206947687,
+    "producer_batches_per_second": 19.123410521882914,
+    "gpu_batches_per_second": 1.0132764335004179,
+    "producer_headroom": 18.8728464312745,
     "sustainable": true,
-    "buffer_seconds": 60.42411827124614,
+    "buffer_seconds": 60.20074876238089,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314151.2848883362,
-    "upload_bytes_per_second": 5430523.647256928,
+    "generation_bytes_per_second": 314164.3032873032,
+    "upload_bytes_per_second": 5623462.468275644,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.40955266940655,
+    "estimated_full_shards_per_hour": 67.41234611477205,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 4826131,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.7090945019999708,
+      "bytes": 13118107,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -45,3 +45,4 @@
 {"step": 359500, "loss": 0.5957230180501938, "lr": 0.0003, "grad_norm": 0.16041565634478688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52621131080, "tokens_per_second": 148511.89742624585, "data_wait_s": 1.3359999684325885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 359600, "loss": 0.5848836395889521, "lr": 0.0003, "grad_norm": 0.16812253405533067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52635588128, "tokens_per_second": 146823.88496027858, "data_wait_s": 9.801000487641431e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 359700, "loss": 0.5889822412282228, "lr": 0.0003, "grad_norm": 0.1657767608538852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52650059176, "tokens_per_second": 146595.91290202373, "data_wait_s": 9.807999958866276e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 359800, "loss": 0.5995389968156815, "lr": 0.0003, "grad_norm": 0.15052040963303098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52664547186, "tokens_per_second": 147907.74579080864, "data_wait_s": 1.2000999959127512e-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 sha256:7b55e3c4239b09ff08517256f3e9e1df5b43f7448151ea5ff8f126adf2a9ce15
-size 99711
+oid sha256:f54b4ea6dd03cd16d1ef9eb0c751ee181266d23a8ea07e6a27f19ec3dd4009f6
+size 101829