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": 77470.30643506463,
-      "required_pages_per_second_upper_estimate": 0.0053595755550542275
+      "required_token_positions_per_second": 76036.79493297855,
+      "required_pages_per_second_upper_estimate": 0.005260401903134966
     },
     {
       "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": 9683.788304383079,
-      "required_pages_per_second_upper_estimate": 0.0040543999425501406
+      "required_token_positions_per_second": 9504.599366622318,
+      "required_pages_per_second_upper_estimate": 0.0039793772762002355
     },
     {
       "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": 19367.576608766158,
-      "required_pages_per_second_upper_estimate": 0.004033755450263874
+      "required_token_positions_per_second": 19009.198733244637,
+      "required_pages_per_second_upper_estimate": 0.003959114789852875
     },
     {
       "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": 29051.364913149235,
-      "required_pages_per_second_upper_estimate": 0.001741584967677536
+      "required_token_positions_per_second": 28513.79809986696,
+      "required_pages_per_second_upper_estimate": 0.0017093586580382604
     },
     {
       "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": 29051.364913149235,
-      "required_pages_per_second_upper_estimate": 0.0022586158385476712
+      "required_token_positions_per_second": 28513.79809986696,
+      "required_pages_per_second_upper_estimate": 0.002216822383321497
     },
     {
       "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": 29051.364913149235,
-      "required_pages_per_second_upper_estimate": 0.37263971618051633
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28513.79809986696,
+      "required_pages_per_second_upper_estimate": 0.36574438629400546
     }
   ],
   "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": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08435098999230206,
-    "producer_p95_seconds": 0.20029644399983226,
-    "gpu_update_seconds": 0.9516936668089545,
-    "producer_batches_per_second": 11.85522541100301,
-    "gpu_batches_per_second": 1.0507582795554555,
-    "producer_headroom": 11.282542942244149,
+    "producer_batch_seconds": 0.071884312132795,
+    "producer_p95_seconds": 0.1901760300006572,
+    "gpu_update_seconds": 0.9696358199341042,
+    "producer_batches_per_second": 13.911241136350538,
+    "gpu_batches_per_second": 1.0313150354407898,
+    "producer_headroom": 13.488837705546294,
     "sustainable": true,
-    "buffer_seconds": 60.90839467577309,
+    "buffer_seconds": 60.11742083591446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 224367.07122780077,
-    "upload_bytes_per_second": 5265441.542320603,
+    "generation_bytes_per_second": 224193.46039937614,
+    "upload_bytes_per_second": 5266107.571686765,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.143950487380195,
+    "estimated_full_shards_per_hour": 48.106697645053515,
     "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": 1.0135722189997978,
-      "bytes": 5920798,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.6641503069999999,
       "bytes": 17682755,
@@ -562,6 +556,12 @@
       "bytes": 6163611,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1898623689994565,
+      "bytes": 6904260,
+      "pages": 25,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -110,3 +110,4 @@
 {"step": 451200, "loss": 0.6011506915092468, "lr": 0.0003, "grad_norm": 0.17583250812376677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66041728407, "tokens_per_second": 151538.91851197154, "data_wait_s": 1.710399919829797e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 451300, "loss": 0.5763578414916992, "lr": 0.0003, "grad_norm": 0.1654944522890505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66056209502, "tokens_per_second": 150050.71314386243, "data_wait_s": 8.071001502685249e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 451400, "loss": 0.5817135572433472, "lr": 0.0003, "grad_norm": 0.1786829287973167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66070755868, "tokens_per_second": 152564.28164295197, "data_wait_s": 1.2328999218880199e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 451500, "loss": 0.6366727352142334, "lr": 0.0003, "grad_norm": 0.1784375598853069, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66085495661, "tokens_per_second": 129471.76160053082, "data_wait_s": 1.1808999261120334e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8392a71597da2fe8c0950f18e601a6a42bb12f7d1fafefe26eaece23a2931be
-size 238947
+oid sha256:1f4ea3d7ce2c2d0e8bb9716d65d31b1764b3b37dc9653adecc46cdaf7df4281e
+size 241065